Artificial intelligence (AI) and machine learning (ML) have been utilized in the medical imaging field over the last several years, offering new ways to enhance workflows, improve accuracy, and unlock deeper insights from complex image data. While the field is continually evolving, and applications vary widely depending on the modality, therapeutic area, and regulatory context, this blog offers a snapshot of key trends, promising use cases, and areas of active development.

From Curiosity to Clinical Tool

AI in medical imaging has progressed from theoretical promise to real-world application. Most current tools employ ML or deep learning (DL) to enhance or automate key aspects of the imaging process, including:

• Image acquisition optimization, reducing scan times or radiation dose
• Lesion detection and segmentation, with improved speed and consistency over manual reads
• Automated triage, flagging high-risk cases for prioritized review
• Quantitative decision support, helping clinicians interpret subtle changes over time

These capabilities are increasingly integrated into commercial platforms and clinical workflows, supporting both efficiency and diagnostic precision.

Where AI Is Making an Impact

• Radiology: AI is now augmenting routine radiology workflows. Algorithms from vendors like Aidoc and Gleamer assist in identifying abnormalities in CTs and X-rays (e.g., pulmonary embolisms, fractures), enabling radiologists to focus on complex interpretive tasks.
• Nuclear Medicine & Radiopharmaceuticals: In nuclear imaging, AI is being used to reconstruct PET/SPECT images, improve dosimetry, and quantify tracer uptake—critical functions in radioligand therapy planning.
• Oncology: DL models assist in tumor segmentation, progression tracking, and even therapy response prediction. When integrated with clinical or genomic data, these tools support the broader vision of precision oncology.
• Neuroimaging: AI-driven pattern recognition and longitudinal analysis have shown promise in early detection of neurodegenerative diseases such as Alzheimer’s, improving prognostic accuracy during critical intervention windows.

Breakthroughs in the Making

Recent advances point to deeper integration of AI across imaging pipelines:

• Multi-modal models are combining imaging with lab values, genomics, and EHRs to boost predictive power.
• Foundation models (large, pre-trained neural networks) are being adapted for flexible image analysis across modalities.
• Generative AI, including GANs and diffusion models, are being used to create synthetic training data—a potential solution to the longstanding bottleneck of annotated datasets.

These technologies are increasingly reflected in FDA-cleared products, not just research settings.

Challenges Still to Overcome

Despite significant momentum, several obstacles remain:

• Generalizability: Many models underperform across different scanners, protocols, or patient populations.
• Bias and standardization: Datasets often reflect systemic biases or lack standardized labeling.
• Clinical adoption: Tools must integrate seamlessly into fast-paced workflows and build trust among clinicians.
• Regulatory uncertainty: Adaptive algorithms and black-box models are pushing regulators like the FDA to refine their oversight frameworks for Software as a Medical Device (SaMD).

Historical Perspectives and Ongoing Interest

It’s important to note that interest in computer-aided detection and interpretation is not new. In fact, studies from the early 2000s—such as one by Elizabeth Krupinski in 2003—found that AI offered limited improvement in radiographic interpretation at the time. With this being said, the landscape has changed dramatically over the last two decades.

Several Bracken consultants were early contributors to this space and hold multiple publications and patents within the arena¹. Dr. Colin Miller, for example, co-developed a clinical workflow tool that enhanced vertebral fracture detection using early algorithmic image characterization methods, leading to both a peer-reviewed publication in Spine and a patent for automated spine assessment using digitized radiographic images (PN US2009297012-A1; WO2009154977-A1).

Similarly, Dr. Wil Reinus has authored multiple foundational studies in the 1990s and early 2000s applying neural networks to mammography screening, emergency CT triage, and bone lesion classification—highlighting the diagnostic potential of AI well before it became mainstream (e.g., Academic Radiology, 1997; Investigative Radiology, 1994).

The PRIMAGE initiative has explored the use of big data and deep learning in childhood oncology, while more recent work (e.g., Shen et al., 2019) has validated AI’s performance in diagnostic support. The Radiological Society of North America (RSNA) now has an AI-based radiology certificate available to members, and regularly features AI topics in tutorials, posters, and symposia, reflecting the field’s growing clinical relevance.

The Regulatory Outlook

The FDA is actively evolving its stance on AI/ML-based SaMD, emphasizing:

• Pre-market review and lifecycle monitoring, especially for adaptive algorithms
• Transparency and explainability, especially in high-stakes clinical areas
• Demonstrated clinical validity and utility, which are especially crucial in oncology and nuclear medicine

Looking Ahead: Smarter, Safer, More Integrated

Future directions suggest a more collaborative, context-aware role for AI:

• AI-powered companion diagnostics to guide radioligand or targeted therapies
• Self-supervised learning to unlock patterns from unlabeled data
• Collaborative AI that augments (not replaces) human decision-making

As these systems mature, their role in delivering precision medicine will only expand, helping ensure that the right diagnosis and treatment reach the right patient, faster and more confidently.

Final Thoughts

AI is revolutionizing medical imaging—but not by replacing radiologists or nuclear medicine specialists. Instead, it’s empowering them to work smarter, faster, and with more data-driven confidence.

As science, regulation, and clinical practice continue to converge, cross-disciplinary expertise will be key to unlocking AI’s full potential in this space.

Looking for support on a medical imaging project? Contact Bracken today to work with our team of expert consultants.

¹Key Patents and Publications 

Colin G. Miller, PhD, FIPEM, CSci:

• Development of a clinical workflow tool to enhance the detection of vertebral fractures. Brett A, Miller CG, Hayes CW, Krasnow J, Ozanian T, Abrams K, Block J, van Kuijk C. Spine 2009;34(22):2437-43.
• Digitized image e.g. x-ray image, characterizing method for mammalian spine, involves automatically characterizing target vertebra based on model parameters during runtime phase, and outputting characterization of vertebra on display device: PN US2009297012-A1; WO2009154977-A1 AE Brett A; Haslam J; Hayes C, Krasnow J, Miller C, van Kuijk C.

William Reinus, MBA, MD

• Reinus WR, Kwasny S, Kalman B.  Neural network-based methods and systems for analyzing complex data. Issued September 4, 2001, #6,285,992.
• Kalman BL, Reinus WR, Kwasny SC, Laine A, Kotner L.  Prescreening entire mammograms for masses using artificial neural networks:  Preliminary results.  Academic Radiology 1997; 4:405-414.
• Reinus WR, Wilson A, Kwasny S, Kalman B.  Diagnosis of focal bone lesions using neural networks.  Investigative Radiology 1994; 29:606-611. 
• Reinus WR, Kalman B, Kwasny S.  Artificial neural networks for screening patients needing emergency cranial computed tomography scans in the emergency departments.  Academic Radiology 1995; 2: 193-198. 
• Kalman BL, Reinus WR, Kwasny SC, Laine A, Kotner L.  Prescreening entire mammograms for masses using artificial neural networks:  Preliminary results.  Academic Radiology 1997; 4:405-414.
• Reinus WR, Wilson A, Kwasny S, Kalman B.  Neural Networks: A new tool for image analysis and application to diagnosis of focal bone lesions.  Marilyn Fixman Cancer Conference, January 1993.
• Reinus WR, Wilson AJ, Kalman B, Kwasny S.  Diagnosis of focal bone lesions using neural networks.  Society of Computer Applications in Radiology. June 1994.
• Reinus WR, Wilson AJ, Kalman B, Kwasny S.  Diagnosis of focal bone lesions using neural networks.  The Association of University Radiologists.  May 1994.
• Reinus WR, Wilson AJ, Kalman B, Kwasny S.  Diagnosis of focal bone lesions using neural networks.  Eighteenth Annual Winter Skeletal Symposium: Musculoskeletal MRI in Telluride, February, 1995.
• Reinus WR, Wippold FJ, Kwasny S, Kalman BJ.  Prediction of the Results of Head CT from Clinical Assessment in Emergency Department Patients:  Univariate and Multivariate Analyses.  Grand Rounds Northwestern Department of Radiology, September, 1997.
• Kalman BL, Kwasny SC, Reinus WR. Classification of subsymbolic data via automatic extraction of statistical features.  Washington University, January, 2000.