The latest advances in AI for MRI radiology

This number best illustrates AI trends in healthcare: among more than 1,000 AI algorithms cleared by the U.S. Food and Drug Administration, 777 are used in radiology, followed by 104 in cardiology. In just five years, the number of scientific studies on AI in radiology has tripled. In 2019, 1,125 scientific studies on this topic were published in PubMed, the biomedical and life sciences literature library. By 2024, this number had risen to 3,804.

For radiologists, this comes as no surprise; their field has always been one of the most innovative. However, much has changed since AI was first applied in the early ’90s. Today, generative AI can synthesize information from medical images, while multimodal AI models combine different types of data—MRI images, electronic medical records, genomics, and other omics—to improve diagnosis.

AI helps radiologists handle repetitive tasks, improve diagnostic accuracy, reduce scan times, and optimize workflows. It plays a crucial role in pathology detection, anomaly identification, and image interpretation, helping meet the growing demand for radiological services. However, AI is just an assistant—it’s radiologists who understand the patient’s medical context, connect the dots between different data points, and make the final diagnosis.

AI in radiology has evolved from simple image-processing tools to sophisticated algorithms that assist in diagnosis and treatment planning. Initially, AI was used for basic image recognition, but today, deep learning models analyze vast datasets to detect minute abnormalities, allowing for early disease diagnosis. AI tools are being integrated into cloud-based radiology platforms and MRI machines, ensuring high data security standards.

“AI can speed up reporting processes and improve workflow efficiency. For a clinic, this means better quality of care, greater precision, and shorter diagnostic times. For patients, it means a shorter time from first symptoms to diagnosis,” says Giacomo Pedretti, an expert in MRI technology at Esaote.

One of AI’s primary contributions to radiology is improving image quality. Advanced AI algorithms reduce noise in MRI scans, leading to sharper images. “Improving image quality means increasing resolution while minimizing noise, which helps radiologists better detect pathologies,” explains Luca Dodero, a specialist in radiology technology at Esaote. This enhanced clarity allows radiologists to make more confident diagnoses and reduces the need for repeat scans. Faster scans also improve patient comfort, particularly for those who experience anxiety during MRI examinations.

Shortening scan times allows more patients to be examined daily, reducing waiting lists and increasing hospital efficiency. Even small time savings have a macro-level impact, improving overall accessibility to medical services. This may seem like a minor adjustment, but when applied to thousands and millions of scans, it brings concrete savings to the healthcare system.

By analyzing thousands of images, AI highlights areas of attention, allowing radiologists to focus on the most relevant findings.

Although AI is widely used in X-ray and CT scans, its role in MRI primarily supports radiologists rather than automating diagnoses. “MRI radiologists still have the final say, but AI assists by streamlining image interpretation and reducing administrative burden,” Pedretti points out.

Beyond imaging, AI is improving radiology workflows. While current AI applications focus on improving image quality, future developments aim to automate repetitive tasks, such as generating preliminary reports and organizing patient data.

Some hospitals and clinics use external cloud AI solutions that process images remotely, while others prefer AI integrated directly into MRI scanners. Integrated AI processes raw imaging data within the device, eliminating the need to send sensitive patient information over the Internet. When embedded inside the scanner, AI allows users to work with raw data rather than filtered images. This ensures higher precision, better compliance with data privacy regulations, and mitigates data security risks.

Cloud-based AI offers flexibility, especially for older MRI machines that lack built-in AI capabilities. However, concerns over data security and GDPR compliance make some institutions hesitant to adopt cloud solutions.

Although generative AI has made it to the covers of medical magazines, its impact on radiology is progressing more slowly than expected. By predicting and reconstructing images from partial data, it has the potential to shorten scan times without compromising diagnostic accuracy. For example, generative AI can reduce the volume of data needed to produce high-quality images, making MRI examinations faster and more efficient.

Just as generative AI assists physicians in making notes and documenting patient records, future technology will enable radiologists to interact more naturally with imaging. Instead of manual workflows requiring clicks or text input, radiologists will be able to “ask” AI to edit an image or display specific layers. The 2023 paper 'Towards Generalist Biomedical AI,' based on experiments with the Med-PaLM Multimodal model developed by Google, provides a good sneak peek into this technology.

However, limitations still stand in the way: AI-generated hallucinations can be misleading, and algorithms' “black box” nature makes it difficult to trace their reasoning. When health and lives are at stake, the technology must be accurate and reliable. If AI is trusted, healthcare professionals will use it for decision-making.

Hospitals and clinics must also ensure that patient data remains protected from cyber threats. “Many healthcare providers are reluctant to put their systems online due to concerns over cyberattacks,” Dodero states.

AI is a transformative technology in radiology. However, one thing is clear: radiologists will be needed more than ever because of the complexity of their job and the growing demand for healthcare services due to an aging population. While AI excels in image analysis, it often lacks the clinical context and nuanced understanding required for accurate diagnosis. Radiologists integrate medical records, physical examinations, and laboratory results to provide comprehensive assessments—something AI alone cannot achieve.

Radiology will remain a human domain, but it will continuously be technology-augmented.

Artur Olesch, Digital Health Journalist, Founder & Editor-in-Chief
aboutdigitalhealth.com