The Rise of AI in Medical Imaging: Transforming Diagnosis

Acknowledgment of Sources and Purpose

This document is a synthesis of ideas, insights, and findings from various authors, researchers, and industry leaders in the field of medical imaging. You can find the authors in the foot notes.  It integrates personal experiences, literature reviews, and advanced technologies discussed in academic articles and professional publications. The purpose is not to claim originality but to compile and contextualize the advancements found in several articles of AI-driven medical imaging, for a broader understanding of my medical colleagues.

Full acknowledgment is given to the original authors whose groundbreaking work inspired this synthesis. All references are cited wherever applicable to ensure due credit is attributed.

1. Introduction

During my time as a medical student, I had the privilege of learning radiology from Dr. Vernor Plaja, a distinguished radiology professor who provided foundational insights into the science and practice of medical imaging.  His lectures were a rigorous combination of theory and hands-on learning, instilling in us a deep understanding of imaging techniques that remain relevant to this day.

One of the most memorable aspects of Dr. Plaja’s teaching was his detailed explanation of X-ray interpretation. He guided us through the nuances of radiographic images, illustrating how variations in density, contrast, and tissue composition could reveal critical diagnostic information. His approach emphasized the importance of observing not just the obvious, but also the subtle details in radiographs, such as small fractures or early signs of pathology, which could easily be overlooked by the untrained eye.

I recall how he encouraged us to apply theoretical concepts in real-world scenarios, often using live demonstrations to show how the imaging department could help us to visualize physiological processes in patients. These interactive sessions allowed us to see how medical technology could provide valuable insights into patient conditions in real-time.

Despite the absence of Magnetic Resonance Imaging (MRI) machines in Costa Rica at the time, Dr. Plaja’s enthusiasm for this modality was evident. He told us about the potential of magnetic resonance imaging (MRI) to provide detailed images of soft tissues, emphasizing its non-invasive nature and ability to avoid the risks associated with ionizing radiation. His descriptions painted a picture of a future where MRI could complement other imaging techniques, offering unprecedented diagnostic clarity, especially for soft tissue pathologies.

Looking back nearly three decades, it is remarkable to reflect on the significant advancements that have since occurred in medical imaging. At that time, it was already evident that the demand for imaging would grow substantially, driven by an increasing patient population and expanding clinical applications. Today, medical imaging has entered an era of rapid transformation, with innovations such as artificial intelligence (AI), machine learning (ML), and enhanced imaging modalities promising to revolutionize diagnostic accuracy and workflow efficiency.

The volume of research required to support these developments, both in public and private healthcare systems, has grown exponentially. The integration of advanced technologies in medical imaging not only promises to improve diagnostic precision but also to address the increasing demand for timely and accurate results in clinical practice.

As we explore the future of medical imaging, the lessons I learned from Dr. Plaja continue to inform my understanding of how far the field has come and the immense potential that lies ahead. The advances in imaging technology—once thought to be the realm of science fiction—are now becoming integral components of modern healthcare, driving improved patient outcomes and redefining the role of medical imaging in clinical decision-making.

The usage of CT scans has seen significant growth, rising from approximately 26 million procedures in 1998 to 62 million by 2006, peaking at 85 million in 2010, and stabilizing at around 74 million per year through 2016. This corresponds to roughly 230 CT procedures per 1,000 people, with common focuses being the abdomen, brain, and chest[1]

These traditional methods accounted for about 281 million procedures in 2006 but dropped slightly to 275 million by 2016, even as the U.S. population grew during this period.[2]  Unfortunately, there is not an accurate number of studies in Latin America that can be compared to the U.S studies.

Overall, radiology analyses remain a substantial aspect of healthcare globally, supported by technological advancements and an increased aging population. The precise annual number of all diagnostic imaging procedures combined, including CT, MRI, X-ray, and nuclear imaging, exceeds several hundred million in the world.

One article from Carestream Health[3], discusses significant challenges faced by radiology services in Latin America, highlighting a growing need for radiologists. The COVID-19 pandemic amplified the demand for fast, secure radiological services and accelerated the adoption of digital imaging technologies. Despite these advancements, radiology departments face difficulties related to increased workloads and staffing shortages, stressing the critical need for more professionals to handle rising patient volumes[4].

Additionally, remote interpretation and mobile radiography have gained importance, emphasizing the necessity for well-trained radiologists to meet new workflow demands and ensure quality patient care.

There is a strong statistical correlation between the number of radiological studies and the number of radiologists globally.  One study identified that the number of images interpreted by each radiologist per year at their health f the US service increased from 467,177 in 2005 to 2,622,176 in 2020, an increase of 561%[5].

As the demand for radiological studies increases, so does the need for radiologists to interpret those images[6]. However, the exact correlation can vary depending on factors such as technological advancements, healthcare system structure, and regional variations in healthcare access.

The rising prevalence of chronic diseases, aging populations, and advancements in medical imaging technology have led to a significant increase in the number of radiological studies performed worldwide.  Despite the growing demand, there is a global shortage of radiologists[7]. This shortage is due to various factors, including limited training programs, aging radiologist populations, and the increasing complexity of imaging studies[8].

In the case of Costa Rica, the lack of specialists in several branches of medicine is due to the poor management of the professional training entity for more than 30 years.  By April 2024, according to the directors of the Costa Rican Social Security Fund (CCSS), the shortage of medical specialists in the country was about 700 doctors[9], and the lack of specialists in radiology and medical imaging was 99 (14% of the total).[10]

While technology has improved image quality and efficiency, it has also increased the volume of images that need to be interpreted. Additionally, newer imaging modalities like PET scans and MRI require specialized expertise as sub specialties in radiology[11].

The availability of radiologists and the demand for radiological studies, can vary significantly between countries and regions. Developed countries with advanced healthcare systems often have a higher demand for radiological services, while developing countries like Costa Rica, may face challenges in attracting and retaining radiologists.

I have several radiologist fiends all over LATAM region, and they confirmed the healthcare system is in need of redefining the strategy by which radiologists are trained globally, in order to meet the demand of more professionals in medical imaging globally.

To address the growing demand for radiological services, several strategies are being explored:

• Expanding Radiologist Training Programs: Increasing the number of radiologist training programs can help alleviate the shortage in the long term[12].
• Utilizing telemedicine and teleradiology can help distribute the workload among radiologists and improve access to imaging services in underserved areas[13].
• AI-powered tools can assist radiologists in image analysis, interpretation, and reporting, increasing efficiency and reducing workload[14].  
• International collaboration with radiologists in other countries can help share expertise and resources, particularly in regions with limited access to specialists.

We all know that medical imaging plays an important role in modern healthcare, helping us clinicians, in diagnosing and monitoring several medical conditions. Technologies such as X-rays, Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and ultrasound, have become indispensable tools for helping doctors to understand the patient’s illness. The increasing demand for more complexity systems and medical data in real time, have led to challenges in the field.

As an example, radiologists are often required to process vast amounts of data under time pressure, potentially leading to diagnostic errors[15].  Artificial intelligence has emerged as a transformative tool in medical imaging.    AI, particularly deep learning and convolutional neural networks (CNNs), is being utilized to assist radiologists by automating image analysis, improving diagnostic accuracy, and reducing time spent on routine tasks[16]. The integration of AI into imaging workflows is addressing some of the key challenges faced by the medical community, including the growing volume of imaging data and the shortage of radiologists in many regions.

I found interesting that AI’s capabilities in medical imaging extend far beyond traditional image processing techniques. One example is AI’s ability to identify patterns in imaging data that are not easily recognizable by human eyes. In a study conducted by McKinney et al[17], Google’s AI system demonstrated superior performance in detecting breast cancer from mammograms compared to human radiologists, with reduced false positives and negatives. The system was trained on thousands of labeled images, enabling it to distinguish malignant tumors from benign masses with high accuracy. This example highlights AI’s potential to not only assist but also enhance the decision-making process in medical imaging.

AI’s adoption can be used in pathologies like Alzheimer’s disease, who can be detected earlier through AI algorithms that analyze MRI scans for subtle changes in brain structure. These models can identify early-stage markers of cognitive decline, offering a critical advantage in managing and treating neurodegenerative diseases[18].

Furthermore, AI’s ability to integrate imaging data with clinical records and genetic information can lead to more personalized treatment plans, marking a shift toward precision medicine.

While AI’s role in medical imaging is still evolving, its potential to revolutionize diagnostic practices is undeniable. The introduction of AI in medical imaging is a significant step toward creating more efficient, accurate, and accessible healthcare systems.

• How AI Works in Medical Imaging

As mentioned before in this and in previous papers, one of its most impactful applications of AI lies in medical imaging, revolutionizing the way medical professionals diagnose and treat diseases. By leveraging advanced algorithms and vast datasets, AI is enabling extraordinary accuracy, speed, and personalization in medical image analysis.

Deep learning technologies can be divided into several families, largely depending on the dominant technological feature. The first point to know is that deep learning gets its name because its networks have many layers of neurons, just like our brain.

For instance, A Fully Connected (FC) layer, (aka a dense layer), is a type of layer used in artificial neural networks where each neuron or node from the previous layer is connected to each neuron of the current layer. It’s called “fully connected” because of this complete linkage.    FC layers are typically found towards the end of neural network architecture and are responsible for producing final output predictions[19]. 

The FC layer takes the high-level features extracted from radiological images (e.g., patterns, edges, shapes) by earlier layers in the neural network and translates them into a decision or classification[20]. For example, in detecting lung nodules on a CT scan, the FC layer processes the abstracted features to determine the likelihood of malignancy.

In breast cancer screening using mammography, FC layers contribute to identifying whether a lesion is benign or malignant by connecting learned features (e.g., density, borders, texture) to known diagnostic categories[21].

One of the things I find intriguing, is that in some ways, the FC layer resembles the cognitive process of a radiologist.  It integrates extracted patterns and contextual information to form a diagnosis. Just as a radiologist considers features such as size, density, and anatomical context, the FC layer combines numerical representations of these features into a unified output, such as a diagnosis or risk score[22].

Something to have in mind, is the fact that the decisions made by FC layers, as mentioned before, can be influenced by biases present in the training data such as overrepresentation of certain pathologies or imaging protocols. For instance, an AI model trained primarily on a specific population might misclassify findings in underrepresented groups[23].

On the other hand, Convolutional Neural Networks (CNNs), begin with several layers that perform convolutions on the input, each of them often followed by pooling layers that combine features while reducing the resolution of an image. Recurrent neural networks (RNNs) have a connection from a later layer to an earlier layer in the network, accounting for the “recurrent” nature pointed to in the name, and these are often applied to situations where the input data is repetitive[24]. 

CNNs were extensively used in many aspects of medical image analysis, allowing for great progress in computer-aided diagnosis in recent years.

At the core of AI’s success in medical imaging is deep learning, a subset of machine learning that involves training neural networks on massive datasets of labeled images[25].

In radiology, deep learning algorithms have demonstrated exceptional performance in differentiating between benign and malignant tumors, aiding in early cancer detection[26].

Medical imaging techniques such as CT Scans, MRI, and Positron Emission Tomography (PET), play a pivotal role in providing us clinicians with detailed and comprehensive visual information about the human body. These imaging modalities generate vast amounts of data that require efficient analysis and interpretation, and this is where AI steps in.

• Health data sources and types

What types of data are useful for AI in medicine?   All health-related data are potentially useful. This includes demographic data, past medical history, family history and social history, clinical examination, laboratory tests and genomic data, and imaging and histopathology, along with lifestyle information, such as nutrition and exercise. I recommend you read this section of the Book, “Artificial Intelligence in Medicine” by Lei Xing Et Al[27], in which they explain in details the various sources of data for a proper validation by AI software.

Increasingly, mobile and wearable technology is proving to be a rich and valuable source of health data.  Crucially, connected devices such as smartphones and watches can be used both to monitor health (e.g., heart rate and rhythm) and to promote healthy behavior, for example, by setting exercise goals.

AI, particularly deep learning algorithms, has demonstrated remarkable capabilities in extracting valuable insights from medical images[28]. Deep learning models, trained on large datasets, can recognize complex patterns and features that may not be readily discernible to the human eye (at least not mine) [29]. These algorithms can even provide a new perspective on what image features should be valued to support clinical decisions [30]. One of the key advantages of AI in medical imaging is its ability to enhance the accuracy and efficiency of disease diagnosis. Through this process, AI can assist healthcare professionals in detecting abnormalities, identifying specific structures, and predicting disease outcomes [31].

By leveraging machine learning algorithms, AI systems can analyze medical images with speed and precision, aiding in the identification of early-stage diseases that may be difficult to detect through traditional methods[32]. This early detection is crucial as it can lead to timely interventions, potentially saving lives and improving treatment outcomes. The incorporation of AI tools into imaging workflows not only expedites the diagnostic process but also alleviates the workload of radiologists, allowing them to concentrate on more complex cases that require human expertise.

Furthermore, AI has opened up new possibilities in image segmentation and quantification. By employing sophisticated algorithms, AI can accurately delineate structures of interest within medical images, such as tumors, blood vessels, or cells [33]. This segmentation capability is invaluable in treatment planning, as it enables clinicians to precisely target areas for intervention, optimize surgical procedures, and deliver targeted therapies [34]. For example, in oncology, precise segmentation of tumors can inform radiation therapy planning, ensuring that healthy tissues are preserved while maximizing the therapeutic effect on cancerous cells[35].

The integration of AI and medical imaging has also facilitated the development of personalized medicine. Through the analysis of medical images and patient data, AI algorithms can generate patient-specific insights, enabling tailored treatment plans that consider individual variations in anatomy, physiology, and disease characteristics[36]. This personalized approach to healthcare enhances treatment efficacy and minimizes the risk of adverse effects, leading to improved patient outcomes and quality of life. Moreover, AI’s predictive capabilities can help identify patients who may benefit most from specific interventions, further refining treatment strategies[37].

Additionally, AI has paved the way for advancements in image-guided interventions and surgical procedures. By combining preoperative imaging data with real-time imaging during surgery, AI algorithms can provide surgeons with augmented visualization, navigation assistance, and decision support[38]. These tools enhance surgical precision, reduce procedural risks, and enable minimally invasive techniques, ultimately improving patient safety and surgical outcomes. As the technology continues to evolve, its applications in intraoperative settings promise to further revolutionize the field of surgery.

The field of medical imaging and AI has experienced significant growth in scholarly publications over recent years. A quick analysis covering the period from 2000 to 2018 identified approximately 8,813 AI-related publications within radiology worldwide, with an exponential increase observed during that timeframe[39].

Further studies indicate that the number of publications on AI in medical imaging rose from about 100–150 per year in 2007–2008 to 700–800 per year in 2016–2017[40].

Additionally, one PUBMED research on artificial intelligence and radiomics in radiology, nuclear medicine, and medical imaging, has shown a continuous growth rate of 26.1%, with an annual growth rate of 29.8%, and a doubling time of approximately 2.7 years[41].

These statistics highlight the rapidly expanding body of literature at the intersection of medical imaging and AI, reflecting the increasing interest and advancements in this interdisciplinary domain.

I have talked in several forums for over 3 years now, that AI has emerged as a transformative force in medical imaging, offering advanced capabilities for image analysis, interpretation, and diagnosis. AI in this field primarily relies on machine learning and deep learning algorithms, which are subsets of AI designed to learn patterns from large datasets, such as medical images. These algorithms process complex information far beyond human visual capabilities, enabling more accurate and efficient diagnostic processes [42].

1. Data Acquisition and Preprocessing

The first step in AI integration into medical imaging involves acquiring high-quality images. These can include X-rays, MRI (Magnetic Resonance Imaging), CT (Computed Tomography), and ultrasound images. The quality of these images is crucial, as AI algorithms learn from the data, they are fed. High-resolution images improve the model’s ability to detect subtle abnormalities.

Preprocessing involves several steps, including normalization (adjusting the brightness and contrast), denoising (removing noise artifacts), and segmentation (isolating regions of interest within the images). These steps are vital to enhance the quality of the input data and ensure that AI algorithms can effectively learn and make accurate predictions[43].

• Model Development

Once the data is preprocessed, it is used to train AI models. Deep learning models, particularly convolutional neural networks (CNNs), have shown significant promise in medical imaging tasks[44]. CNNs are designed to automatically detect features in images, such as edges, shapes, and textures, which are crucial for recognizing patterns associated with various diseases.

The training process involves feeding the model large datasets of labeled images, where the labels indicate the presence or absence of specific conditions (e.g., tumors, fractures). The model iteratively adjusts its internal parameters to minimize the difference between its predictions and the actual labels. This process, known as supervised learning, relies heavily on the quality and quantity of training data[45].

• Evaluation and Validation

After training, the model must be evaluated using separate validation datasets[46]. This ensures that the model can generalize its learning to new, unseen data. Metrics such as accuracy, sensitivity, specificity, and area under the curve (AUC) are commonly used to assess model performance[47]. Cross-validation techniques can help to further ensure that the model is robust and reliable.

It’s essential for us medical practitioners, to trust AI’s outputs; thus, extensive validation against standard diagnostic practices is crucial. For example, a study by Esteva et al[48] demonstrated that a deep learning model could classify skin cancer with a performance comparable to that of experienced dermatologists.

• Deployment in Clinical Settings

Once validated, AI models can be deployed in clinical settings. They assist radiologists in interpreting medical images, highlighting areas of concern, and even providing preliminary diagnoses[49]. For instance, AI algorithms can analyze chest X-rays to detect pneumonia or nodules indicative of lung cancer, thereby improving the speed and accuracy of radiological assessments [50].

Integration into clinical workflows is critical. AI tools should complement human expertise rather than replace it. Radiologists can utilize AI-generated insights to enhance their decision-making, reduce workload, and improve patient outcomes.

• Challenges and Ethical Considerations

Despite its potential, the use of AI in medical imaging presents several challenges. These include data privacy concerns, algorithmic bias, and the need for continuous training with diverse datasets to ensure generalizability across populations. Moreover, the interpretability of AI models remains a concern; understanding how an AI arrived at a particular conclusion is crucial for clinical acceptance[51].

Additionally, regulatory frameworks must adapt to accommodate AI technologies in healthcare, ensuring that they meet safety and efficacy standards before widespread adoption.

• Key Applications of AI in Medical Imaging

Artificial intelligence has emerged as a key player in medical imaging, providing solutions to some of the most pressing challenges faced by radiologists and clinicians. By leveraging AI, healthcare systems are not only improving the accuracy of diagnoses but also making medical imaging more efficient and accessible We will explore the different applications of AI across various imaging modalities, while also discussing real-world implementations and case studies that demonstrate the transformative power of AI in this field.

1. AI in Conventional Radio Imaging

Radiography, including X-rays, has long been the cornerstone of diagnostic imaging. Despite being one of the oldest imaging techniques, it continues to play a vital role in the initial assessment of fractures, chest abnormalities, and more. AI has begun enhancing the traditional radiography workflow by automating image analysis and triaging. For instance, CheXNet, a deep learning algorithm developed by Stanford University, was trained to diagnose pneumonia from chest X-rays, outperforming radiologists in some cases [52]. The system demonstrated high sensitivity and specificity, particularly in distinguishing between various lung pathologies, showing the potential to aid in early detection and intervention.

Beyond pneumonia, AI applications in radiography extend to musculoskeletal injuries, lung cancer screening, and cardiovascular abnormalities. AI algorithms can rapidly process X-rays, identifying abnormalities that might be missed due to human error or fatigue[53]. Soon, I foresee that all radiology departments in the world will have some type of AI tool in their department that become a routine part of radiography workflow, allowing radiologists to focus on more complex cases while AI handles preliminary analysis.

1. AI in Magnetic Resonance Imaging (MRI)

AI is significantly enhancing MRI imaging by improving image quality, reducing scan times, and aiding in more accurate diagnoses. Several AI-based MRI software solutions have gained prominence in the medical imaging field.

AI-powered MRI solutions have transformed diagnostic radiology by enhancing image quality, optimizing workflow efficiency, and aiding in early and precise disease detection. These solutions utilize machine learning algorithms to improve signal-to-noise ratios, denoise images, and accelerate scan times, thereby reducing patient discomfort and increasing throughput in radiology departments[54]. For example, SubtleMR™ by Subtle Medical uses AI to refine MRI images, ensuring clarity and diagnostic accuracy even with shorter scan durations or lower contrast agent dosages, which can benefit patients with renal impairment or contrast allergies[55]

One critical application of AI in MRI is in neuroimaging, where it aids in detecting and monitoring neurological disorders. AI algorithms can identify structural changes in brain imaging that are characteristic of diseases such as Alzheimer’s, multiple sclerosis, and glioblastoma.[56] Tools like icobrain, developed by icometrix, analyze volumetric MRI data to track disease progression in neurodegenerative conditions, providing radiologists and neurologists with actionable insights. Moreover, AiMIFY enhances contrast in brain MRI scans, enabling the detection of subtle abnormalities with lower doses of contrast agents, improving safety and diagnostic yield for patients[57].

AI-powered MRI solutions are also pivotal in musculoskeletal imaging. They enable the precise identification of soft tissue injuries and degenerative changes in cartilage, ligaments, and tendons. AI platforms such as Aidoc can detect abnormalities like meniscal tears or early osteoarthritis with high sensitivity, assisting orthopedic surgeons in treatment planning[58]. Additionally, AI’s ability to detect incidental findings and quantify changes over time is invaluable in longitudinal studies, improving patient care and advancing clinical research[59]. These capabilities underscore the transformative role of AI in modern MRI diagnostics, enhancing accuracy, efficiency, and patient safety.

Another commercial solution is AiMIFY, an AI-powered software designed to enhance contrast in brain MRI scans, offering a groundbreaking solution for more precise imaging. By leveraging artificial intelligence, the software achieves significantly improved contrast at reduced dosages of contrast agents, effectively doubling the diagnostic quality of traditional methods[60]. This innovation enhances the detection of abnormalities, particularly in neurological conditions, providing clinicians with clearer, more actionable data while minimizing patient exposure to contrast agents. The integration of AI not only optimizes imaging protocols but also streamlines workflows, enabling healthcare professionals to make faster and more accurate decisions, ultimately improving patient outcomes.

• AI in Computed Tomography (CT SCAN)

In every radiology department in modern hospitals, the CT SCAN is the workhorse of this service.  In Latin America, this diagnostic method became the doctor’s method of choice to diagnose several pathologies. 

The benefits of CT allow physicians for more effective medical care by:

• Determining when surgeries are needed
• Reducing the need for exploratory surgeries
• Improving cancer diagnosis and treatment
• Reducing hospitalization time
• Guiding the treatment of common conditions such as injuries, heart disease, and stroke
• Improving patient placement to appropriate areas of care, such as intensive care units

In the emergency room, patients can be scanned quickly so that doctors can quickly assess the patient’s condition.

AI-powered solutions for CT imaging are revolutionizing diagnostic radiology by improving image quality, reducing radiation exposure, and accelerating workflow efficiency. By leveraging advanced deep learning algorithms, these technologies can denoise low-dose CT scans, preserving diagnostic quality while minimizing patient radiation exposure[61]. For instance, SubtlePET™, developed by Subtle Medical, applies AI to optimize CT images acquired with lower doses of contrast agents, maintaining high-resolution imaging crucial for accurate diagnoses in oncology and cardiology patients[62]These innovations enable healthcare providers to enhance patient safety while maintaining diagnostic integrity.

In pulmonary imaging, AI-powered CT solutions are particularly impactful in detecting and monitoring lung diseases, including COVID-19 pneumonia and pulmonary fibrosis. Algorithms such as those developed by Aidoc identify subtle changes in lung parenchyma with high sensitivity, enabling early detection and precise quantification of disease progression[63]. Similarly, Zebra Medical Vision’s AI tools analyze chest CT scans to detect incidental findings, such as coronary artery calcifications, osteoporosis, or emphysema, offering clinicians valuable insights that can guide comprehensive patient management.[64] This capability highlights AI’s role in enhancing the utility of CT imaging for both targeted and incidental findings.

AI also facilitates cardiovascular imaging by automating the detection and quantification of atherosclerotic plaques, aneurysms, and other vascular anomalies. HeartFlow, for example, uses AI to create 3D models of coronary arteries from CT angiography scans, enabling non-invasive assessment of fractional flow reserve and guiding treatment decisions in patients with coronary artery disease.[65]  This approach reduces the need for invasive diagnostic procedures and improves patient outcomes. These examples demonstrate how AI-powered CT solutions enhance diagnostic precision, streamline workflows, and contribute to more personalized and effective patient care.

• AI in Ultrasound

Is well known by physicians and patients that ultrasound is a widely used imaging technique due to its safety, cost-effectiveness, and versatility. However, interpreting ultrasound images is highly operator-dependent, leading to variability in diagnostic accuracy.

AI-powered solutions for ultrasound imaging are significantly improving diagnostic capabilities by automating processes, enhancing image quality, and enabling the detection of subtle abnormalities. These solutions utilize deep learning and computer vision algorithms to optimize ultrasound workflows and reduce operator dependency, which has historically been a challenge for this modality[66].  Fone company called Caption Health, has developed AI-based software that guides clinicians in acquiring high-quality cardiac ultrasound images, even without extensive training, democratizing access to echocardiography and improving diagnostic capabilities in remote or resource-limited settings.[67]

Cardiac imaging is one of the primary areas where AI in ultrasound has had a profound impact. AI algorithms are used to automate measurements such as ejection fraction, ventricular size, and wall motion abnormalities, which are critical for diagnosing heart failure and valvular diseases. Tools like GE Healthcare’s Vivid IQ with AI capabilities enhance accuracy and reduce variability in echocardiographic assessments, ensuring more consistent patient management[68].  Moreover, AI solutions can assist all type of physicians (not only ER or cardiologists) in real-time decision-making during emergency scenarios, but such also as detecting pericardial effusions or major cardiac dysfunction, enabling faster treatment interventions.

Another company, Caption Health recently received CE Mark certification for its Caption AI™ technology platform. This platform utilizes artificial intelligence to facilitate the acquisition of diagnostic cardiac ultrasound images by medical staff, regardless of their prior sonography experience, thereby enhancing early detection of cardiac conditions such as heart failure and valvular heart disease.  I have a presentation that I teach in the university about how this product could be a gamechanger to primary care medicine and eliminate the mobilization of patients from rural zones to a bigger hospital by training a General Practitioner or a nurse to use this product, with the secondary support of one specialist to validate the results of the cardiac study made with this product.

Other notable application is in obstetrics and gynecology, where AI-powered ultrasound software aids in detecting fetal abnormalities, estimating gestational age, and evaluating maternal health. For instance, Philips Ultrasound AI solutions offer automated fetal biometry and anomaly detection, which enhance prenatal care by providing consistent and accurate measurements[69]. Additionally, AI-powered systems are used in liver imaging for identifying fibrosis or steatosis and in breast ultrasound for characterizing masses as benign or malignant, reducing unnecessary biopsies[70]. These applications highlight AI’s ability to expand the diagnostic utility of ultrasound across multiple specialties, improving clinical outcomes and operational efficiency.

• Challenges and Limitations of AI-Based Technology in Medical Imaging Devices

AI-based technology in medical imaging presents remarkable opportunities but also faces several challenges and limitations that require careful consideration. One significant challenge is the lack of generalizability and bias in AI models.  I wrote a previous paper on the importance of AI algorithms that are often trained on datasets that may not represent diverse patient populations, leading to reduced accuracy in underrepresented demographics or atypical clinical cases[71]. The «black-box» nature of many AI systems makes it difficult for physicians to understand how decisions are made, creating potential barriers to trust and adoption in clinical practice[72]. Regulatory challenges also pose limitations, as the approval process for AI-driven tools varies significantly between regions, and updates to algorithms often require re-approval.

Another critical limitation is the need for robust integration into existing healthcare infrastructure. AI-powered solutions often require high-performance computing resources, seamless integration with “Picture Archiving and Communication Systems” (PACS), and adherence to data security and privacy regulations such as HIPAA or GDPR[73].  Additionally, physician training is necessary to ensure proper use and interpretation of AI outputs. If misinterpreted, AI-generated results could lead to diagnostic errors or overreliance on technology, potentially compromising patient safety.[74]  These factors contribute to financial and operational challenges in adopting AI technologies at scale.

Physicians and healthcare institutions should evaluate several factors before investing in AI-based medical imaging technology. Firstly, they should assess the clinical validation and regulatory approval status of the AI software, ensuring its performance has been rigorously tested in diverse, real-world[75]. It is also essential to evaluate whether the AI solution aligns with the institution’s clinical needs, such as enhancing diagnostic accuracy in specific modalities or addressing workflow bottlenecks. Cost-effectiveness analysis should be conducted to determine whether the anticipated benefits, such as improved patient outcomes or increased throughput, justify the investment.

Additionally, hospitals must ensure robust data infrastructure and cybersecurity measures to support AI adoption. Interoperability with existing imaging systems and electronic health records is critical for seamless integration. Finally, ongoing education and training programs for clinicians are essential to foster understanding and appropriate use of AI technologies, emphasizing that these tools are meant to augment, not replace, clinical expertise. By addressing these considerations, institutions can maximize the potential benefits of AI while mitigating risks and challenges.

• The Future of AI in Medical Imaging

The future of AI in medical imaging is poised to revolutionize healthcare delivery by enabling earlier disease detection, personalized treatment planning, and more efficient workflows. Advanced AI algorithms are expected to evolve from mere diagnostic tools to predictive and prescriptive systems, capable of identifying disease risk factors and suggesting tailored interventions[76]. For example, deep learning models might predict tumor response to therapy based on imaging biomarkers, aiding oncologists in optimizing treatment regimens[77].  Similarly, AI may facilitate real-time imaging during procedures, such as guiding catheter placement in interventional radiology or optimizing views in echocardiography, making it a critical partner in precision medicine.

Another promising direction is the integration of AI into multi-modal imaging and data fusion. By combining information from CT, MRI, PET, and ultrasound with patient-specific data from electronic health records (EHRs), AI could provide holistic insights that enhance diagnostic accuracy and clinical decision-making[78].  For instance, AI might integrate imaging data with genetic profiles to identify patients at risk for specific conditions, such as cardiovascular disease or cancer, long before clinical symptoms appear. This convergence of imaging and genomics could usher in a new era of preventive medicine and risk stratification.

Lastly, AI’s role in automating routine tasks and managing imaging backlogs is expected to grow. Future AI solutions may take on administrative tasks, such as protocol optimization, scheduling, and report generation, freeing up radiologists to focus on complex cases[79].  Additionally, AI-powered systems might aid in global health by providing low-cost, high-quality diagnostic support in resource-limited settings. With advancements in edge computing, even remote clinics may have access to sophisticated imaging analytics, bridging gaps in healthcare disparities. As these technologies mature, fostering collaboration between AI developers, clinicians, and regulatory bodies will be essential to ensure safe, equitable, and impactful integration of AI into medical imaging.

• Conclusion

Next year it will be my 30th anniversary as a Physician.  In this 3 decades, I have seen the evolution of medicine in the pharmaceutical, clinical, medical devices and hospital environments.   I never thought I could see the things current medicine can do now in those fields.  I believe the integration of AI tools into medical imaging is not merely a technological advancement, but a transformative shift in how we practice medicine.

AI offers unparalleled potential to enhance diagnostic precision, optimize workflows, and personalize patient care, enabling clinicians to deliver more efficient and effective treatments. However, it is essential to recognize that these tools are not a replacement for the expertise and judgment of healthcare professionals. Rather, they serve as powerful allies, augmenting our abilities and allowing us to focus on what truly matters, the human connection with our patients.

The journey toward widespread adoption of AI in medical imaging is not without challenges. Issues such as algorithm bias, lack of transparency, and integration hurdles require careful attention to ensure equitable and safe implementation. As we navigate these complexities, ongoing collaboration between physicians, developers, and regulatory bodies is critical to shaping AI solutions that are clinically meaningful and ethically sound. Moreover, the role of education cannot be overstated; clinicians must be equipped with the skills to critically evaluate and utilize AI outputs while maintaining accountability for clinical decisions.

In my view, the future of AI in medical imaging is one of immense promise. By embracing this technology thoughtfully and responsibly, we could redefine healthcare, making it more accurate, accessible, and patient-centered. As stewards of this transformation, we must ensure that AI serves as a tool to enhance, rather than replace, the art and science of medicine. Together, we can harness the power of AI to elevate the quality of care we provide, ensuring that innovation translates into better outcomes for patients worldwide.

---

[1] Nigatu, A. M., Yilma, T. M., Gezie, L. D., Gebrewold, Y., Gullslett, M. K., Mengiste, S. A., & Tilahun, B. (2023). Medical imaging consultation practices and challenges at public hospitals in the Amhara regional state, Northwest Ethiopia: a descriptive phenomenological study. BMC Health Services Research, 23(1). https://doi.org/10.1186/s12913-023-09652-9

[2] Trends in the number of radiology studies and a decline in radiation dose. (2022, December 19). Health Imaging. https://healthimaging.com/topics/medical-imaging/trends-number-radiology-studies-and-decline-radiation-dose

[3] Warner, M. (2022, February 9). Cubrir las lagunas de la escasez de técnicos radiólogos. Everything Rad. https://www.carestream.com/blog/2022/02/09/cubrir-las-lagunas-de-la-escasez-de-tecnicos-radiologos/

[4] Warner, M. (2021, April 26). Desafíos y tendencias en radiología en América Latina para la segunda mitad de 2021. Everything Rad. https://www.carestream.com/blog/2021/04/24/desafios-y-tendencias-en-radiologia-en-america-latina-para-la-segunda-mitad-de-2021/

[5]Peng Y-C, Lee W-J, Chang Y-C, Chan WP, Chen S-J. Radiologist burnout: trends in medical imaging utilization under the national health insurance system with the universal code bundling strategy in an academic tertiary medical centre. Eur J Radiol 2022; 157: 110596.

[6] Caliber Healthcare Solutions. (n.d.). The growing demand for radiologists: understanding, adapting, and meeting the need. https://www.caliberhealth.com/blog/are-radiologists-in-demand#:~:text=The%20short%20answer%20is%20yes,typically%20requires%20more%20medical%20imaging.

[7] Jeganathan S. The Growing Problem of Radiologist Shortages: Australia and New Zealand’s Perspective. Korean J Radiol. 2023 Nov;24(11):1043-1045. doi: 10.3348/kjr.2023.0831. PMID: 37899512; PMCID: PMC10613841.

[8] Jeganathan S. The Growing Problem of Radiologist Shortages: Australia and New Zealand’s Perspective. Korean J Radiol. 2023 Nov;24(11):1043-1045. doi: 10.3348/kjr.2023.0831. PMID: 37899512; PMCID: PMC10613841.

[9] https://semanariouniversidad.com/pais/ccss-tiene-escasez-de-700-medicos-especialistas-180-son-anestesiologos-y-radiologos-senala-marta-esquivel/

[10] https://semanariouniversidad.com/pais/ccss-tiene-escasez-de-700-medicos-especialistas-180-son-anestesiologos-y-radiologos-senala-marta-esquivel/

[11] Benamor M, Ollivier L, Brisse H, Moulin-Romsee G, Servois V, Neuenschwander S. PET/CT imaging: what radiologists need to know. Cancer Imaging. 2007 Oct 1;7 Spec No A(Special issue A):S95-9. doi: 10.1102/1470-7330.2007.9012. PMID: 17921089; PMCID: PMC2727970.

[12] Radiology facing a global shortage. (n.d.). https://www.rsna.org/news/2022/may/global-radiologist-shortage#:~:text=%E2%80%9CFrom%20a%20training%20perspective%2C%20the,resident%20positions%2C%E2%80%9D%20said%20Dr.

[13] Teleradiology — Can radiologists work from home? | Candelis. (n.d.). https://www.candelis.com/blog/can-radiologists-work-from-home#:~:text=Workload%20balancing%3A%20Teleradiologists%20distribute%20their%20workload%20more%20evenly%2C%20improving%20efficiency.

[14] Teleradiology — Can radiologists work from home? | Candelis. (n.d.). https://www.candelis.com/blog/can-radiologists-work-from-home#:~:text=Workload%20balancing%3A%20Teleradiologists%20distribute%20their%20workload%20more%20evenly%2C%20improving%20efficiency.

[15] Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44-56. https://doi.org/10.1038/s41591-018-0300-7

[16] Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118. https://doi.org/10.1038/nature21056

[17] McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., … & Suleiman, Y. (2020). International evaluation of an AI system for breast cancer screening. Nature, 577(7788), 89-94. https://doi.org/10.1038/s41586-019-1799-6

[18] Boehm, K., Salomon, G., Beyer, B., Schiffmann, J., Simonis, K., & Graefen, M. (2020). Artificial intelligence in prostate cancer imaging: Deep learning in multiparametric MRI. Der Radiologe, 60(12), 1099-1108. https://doi.org/10.1007/s00117-020-00730-x

[19] GeeksforGeeks. (2024, May 18). Fully Connected Layer vs Convolutional Layer. GeeksforGeeks. https://www.geeksforgeeks.org/fully-connected-layer-vs-convolutional-layer/

[20] Yamashita R, Nishio M, Do RKG, Togashi K. Convolutional neural networks: an overview and application in radiology. Insights Imaging. 2018 Aug;9(4):611-629. doi: 10.1007/s13244-018-0639-9. Epub 2018 Jun 22. PMID: 29934920; PMCID: PMC6108980.

[21] Vivek Kumar Singh Et Al.  “Breast tumor segmentation and shape classification in mammograms using generative adversarial and convolutional neural network”. Expert Systems with Applications. Volume 139, January 2020, 112855. https://doi.org/10.1016/j.eswa.2019.112855

[22] Lim S, Radicchi F, van den Heuvel MP, Sporns O. Discordant attributes of structural and functional brain connectivity in a two-layer multiplex network. Sci Rep. 2019 Feb 27;9(1):2885. doi: 10.1038/s41598-019-39243-w. PMID: 30814615; PMCID: PMC6393555.

[23] Kakko, A. (2024, December 26). Understanding dataset bias in artificial intelligence: causes, consequences, and solutions. Alphanome.AI. https://www.alphanome.ai/post/understanding-dataset-bias-in-artificial-intelligence-causes-consequences-and-solutions#:~:text=Dataset%20bias%20occurs%20when%20training%20data%20used%20to,use%20case%20the%20model%20is%20intended%20to%20serve.

[24] LEI XING. ARTIFICIAL INTELLIGENCE IN MEDICINE. Technical Basis and Clinical Applications. Elsevier. ISBN: 978-0-12-821259-2

[25] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.

[26] Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Tenenbaum, J. L., … & Dermatologist, C. (2017). A Dermatologist-Level Classification of Skin Cancer with Deep Neural Networks. Nature, 542(7639), 115-118.

[27] Artificial intelligence in medicine. (n.d.). Elsevier Shop. https://shop.elsevier.com/books/artificial-intelligence-in-medicine/xing/978-0-12-821259-2

[28] Kalloo, S., Iqbal, M. S., & Smith, K. (2023). How Artificial Intelligence Is Shaping Medical Imaging Technology: A Survey of Innovations and Applications. Journal of Medical Imaging, 10(2), 1-20.

[29] Zhang, Y., Li, H., & Liu, J. (2020). Deep Learning for Medical Image Analysis: A Comprehensive Overview. Journal of Biomedical Engineering, 137(2), 020901.

[30] Ge, S., Wang, Y., et al. (2021). Deep Learning for Medical Image Segmentation: A Survey. Medical Image Analysis, 67, 101881.

[31] Kourou, K., Exarchos, T. P., et al. (2015). Machine Learning Applications in Cancer Prognosis and Prediction. Computational and Structural Biotechnology Journal, 13, 8-17.

[32] Chandra, R. (2024, August 1). AI in Medical Image Analysis: Redefining Diagnostic Accuracy and Speed. Daffodil Unthinkable Software Corp. https://insights.daffodilsw.com/blog/ai-in-medical-image-analysis-redefining-diagnostic-accuracy-and-speed

[33] Chen, M., & Wang, Y. (2019). Medical image segmentation using deep learning: a review. Artificial Intelligence in Medicine, 99, 101-118.

[34] Rundo, L., et al. (2018). Advanced Imaging Techniques for Tumor Detection and Characterization: An Overview. Journal of Digital Imaging, 31(3), 303-318.

[35] Ricky R. Savjani, Michael Lauria, Supratik Bose, Jie Deng, Ye Yuan, Vincent Andrearczyk, Automated Tumor Segmentation in Radiotherapy, Seminars in Radiation Oncology, Volume 32, Issue 4, 2022, Pages 319-329, ISSN 1053-4296, https://doi.org/10.1016/j.semradonc.2022.06.002.

[36] Obermeyer, Z., & Emanuel, E. J. (2016). Predicting the Future—Big Data, Machine Learning, and Clinical Medicine. New England Journal of Medicine, 375, 1216-1219.

[37] Tzeng, J. M., et al. (2020). The Role of Artificial Intelligence in Personalized Medicine: A Review. Bioengineering & Translational Medicine, 5(1), e10146.

[38] Yang, G., et al. (2017). Medical Image Computing and Computer-Assisted Intervention: A Review. Artificial Intelligence in Medicine, 75, 25-30

[39] West, E., Mutasa, S., Zhu, Z., & Ha, R. (2019). Global trend in Artificial Intelligence–Based publications in Radiology from 2000 to 2018. American Journal of Roentgenology, 213(6), 1204–1206. https://doi.org/10.2214/ajr.19.21346

[40] Pesapane, F., Codari, M., & Sardanelli, F. (2018). Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine. European Radiology Experimental, 2(1). https://doi.org/10.1186/s41747-018-0061-6

[41] Kocak, B., Baessler, B., Cuocolo, R., Mercaldo, N., & Santos, D. P. D. (2023). Trends and statistics of artificial intelligence and radiomics research in Radiology, Nuclear Medicine, and Medical Imaging: bibliometric analysis. European Radiology, 33(11), 7542–7555. https://doi.org/10.1007/s00330-023-09772-0

[42] Pinto-Coelho L. How Artificial Intelligence Is Shaping Medical Imaging Technology: A Survey of Innovations and Applications. Bioengineering (Basel). 2023 Dec 18;10(12):1435.

[43] Litjens, G., Kooi, T., Bejnordi, B. E., et al. (2017). A survey on deep learning in medical image analysis. Medical Image Analysis, 42, 60-88.

[44] Soffer, S., Ben-Cohen, A., Shimon, O., Amitai, M. M., Greenspan, H., & Klang, E. (2019). Convolutional Neural Networks for Radiologic Images: A Radiologist’s Guide. Radiology, 290(3), 590-606. https://doi.org/10.1148/radiol.2018180547

[45] Shen, D., Wu, G., & Suk, H. I. (2017). Deep Learning in Medical Image Analysis. Annual Review of Biomedical Engineering, 19, 221-248.

[46] Kornilov, A. (2024, 14 mayo). Training, validation, and test datasets. What is the difference? Trainingdata.pro. https://trainingdata.pro/training-validation-test-datasets

[47] GeeksforGeeks. (2024, 27 mayo). Generalization Rules in AI. GeeksforGeeks. https://www.geeksforgeeks.org/generalization-rules-in-ai/

[48] Esteva, A., Kuprel, B., Wang, S., et al. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118.

[49] Medscience, O. (2024, 17 junio). The Role of AI in Radiology Transforming Healthcare. Open MedScience. https://openmedscience.com/the-role-of-ai-in-radiology-transforming-healthcare/

[50] Lakhani, P., & Sundaram, B. (2017). Deep Learning at Scale in Medical Imaging: A Review. Journal of Medical Imaging, 4(4), 041205.

[51] Char, D. S., Shah, N. H., & Magnus, D. (2018). Implementing Machine Learning in Health Care—Addressing Ethical Challenges. The New England Journal of Medicine, 378(11), 981-983

[52] Rajpurkar, P., Irvin, J., Zhu, K., Yang, B., Mehta, H., Duan, T., … & Ng, A. Y. (2017). CheXNet: Radiologist-level pneumonia detection on chest X-rays with deep learning. arXiv preprint arXiv:1711.05225. Retrieved from https://arxiv.org/abs/1711.05225

[53] Chandra, R. (2024, 1 agosto). AI in Medical Image Analysis: Redefining Diagnostic Accuracy and Speed. Daffodil Unthinkable Software Corp. https://insights.daffodilsw.com/blog/ai-in-medical-image-analysis-redefining-diagnostic-accuracy-and-speed

[54] Wang, G., Zhang, S., & Rutt, B. (2021). AI in medical imaging: Applications in MRI reconstruction and analysis. Nature Machine Intelligence, 3(10), 943–952.

[55] Subtle Medical. (2023). SubtleMR: Enhancing MRI image quality with AI. Retrieved from https://subtlemedical.com

[56] Feng, X., Zhang, H., & Dong, Y. (2022). Advances in AI-powered MRI for neurodegenerative diseases: A systematic review. Journal of Neurology, 269(5), 2452–2460.

[57] Icometrix. (n.d.). icometrix. https://icometrix.com/

[58] Aidoc. (2023). AI solutions for musculoskeletal imaging. Retrieved from https://aidoc.com

[59] Glocker, B., Robinson, R., & Castro, D. C. (2020). Machine learning in musculoskeletal imaging: Current status and future opportunities. Radiology: Artificial Intelligence, 2(1), e190028.

[60] Hall, J. (2024, October 16). FDA clears emerging AI-Powered software for brain MRI. Diagnostic Imaging. https://www.diagnosticimaging.com/view/fda-clears-emerging-ai-powered-software-for-brain-mri?utm_source=chatgpt.com

[61] McCollough, C. H., Leng, S., & Yu, L. (2022). Advances in low-dose CT imaging: The role of AI and iterative reconstruction techniques. Radiology, 305(2), 249–259.

[62] Subtle Medical. (2023). SubtlePET and SubtleCT: Optimizing CT imaging with AI. Retrieved from https://subtlemedical.com

[63] Aidoc. (2023). AI solutions for pulmonary imaging. Retrieved from https://aidoc.com

[64] Zebra Medical Vision. (2023). AI tools for incidental findings in CT imaging. Retrieved from https://zebra-med.com

[65] Taylor, C. A., Fonte, T. A., & Min, J. K. (2021). Noninvasive coronary assessment using computed tomography and AI-based modeling: Applications of HeartFlow. Journal of Cardiovascular Computed Tomography, 15(3), 245–252.

[66] Huang, Z., Zhao, J., & Zhou, S. (2021). Artificial intelligence in ultrasound imaging: A review of current applications and future prospects. Medical Physics, 48(8), 3786–3802.

[67] Caption Health. (2023). AI-guided ultrasound imaging solutions. Retrieved from https://captionhealth.com

[68] Bajorek, A., Howard, J. P., & Shun-Shin, M. J. (2022). AI in echocardiography: Current applications and future directions. Journal of the American College of Cardiology: Cardiovascular Imaging, 15(4), 547–560.

[69] Philips. (2023). AI solutions for obstetric and gynecologic ultrasound. Retrieved from https://philips.com

[70] Cho, J., Kim, E. K., & Oh, Y. (2020). AI in breast ultrasound imaging: Early detection and reducing false positives. Ultrasound in Medicine & Biology, 46(5), 1238–1245.

[71] Willemink, M. J., Koszek, W. A., & Hardell, C. (2020). Preparing medical imaging data for machine learning. Radiology, 295(1), 4–15.

[72] Jin, C., Chen, W., & Cao, Y. (2021). AI and bias in radiology: Implications and strategies for improvement. Radiology, 298(2), 321–330.

[73] Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44–56.

[74]Liew, C. (2022). The impact of AI on medical imaging: Challenges and considerations. Insights into Imaging, 13(1), 45.

[75] Davenport, T., & Kalakota, R. (2019). The potential for AI in healthcare. Future Healthcare Journal, 6(2), 94–98.

[76] Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44–56.

[77] Sun, W., Zheng, B., & Qian, W. (2021). Predictive modeling in cancer treatment: AI applications in medical imaging. Journal of Clinical Oncology, 39(5), 556–564.

[78] Davenport, T., & Kalakota, R. (2019). The potential for AI in healthcare. Future Healthcare Journal, 6(2), 94–98.

[79] Tang, A., Tam, R., & Cadrin-Chênevert, A. (2022). The role of artificial intelligence in reducing workload and increasing efficiency in radiology departments. Radiology: Artificial Intelligence, 4(1), e210265.