The Evolution of AI Models in Mammography With Robert Nishikawa, PhD

Studies focused on applying deep-learning artificial intelligence (AI) models to mammography screenings have shown promising results in improving radiologist reading times and reducing false negatives 1-3

On this episode of Managed Care Cast, The American Journal of Managed Care® spoke with Robert Nishikawa, PhD, a professor in the Department of Radiology at the University of Pittsburgh, in light of Breast Cancer Awareness Month. Nishikawa’s career and research focus on breast imaging technology, development, implementation, and assessment.

Nishikawa discussed the practical implications and evolution of AI models used in mammography screenings to aid radiologists’ readings and advance breast cancer screenings for patients.

Recent developments in deep-learning AI models applied to mammography screening improved double-radiologist readings in the Netherlands while detecting a significant number of malignancies. However, when forced to match the high specificity of double-radiologist readings, its sensitivity was far less.1

Contrary to European practices of double-radiologist readings, the US only requires a single reader with the assistance of computer-aided detection.2 Whereas European studies supplement the second radiologist’s readings with deep-learning AI models, practical application is not as feasible for studies in the US, Nishikawa said. However, when used, it allowed radiologists to read screenings faster at a higher sensitivity and specificity, he said.

To further develop the accuracy of deep-learning AI models, Nishikawa suggested using them in clinical settings to gauge their effectiveness in the US. As the AI models cannot replace single-radiologist readings, Nishikawa emphasized it’s up to the radiologist to decide whether they want to use it or not. Within that, he explained that radiologists must maintain a certain level of trust when interacting with AI, as current algorithms are “very accurate.”

Clinical applications of AI models may also optimize survivorship care, as screenings are one of the primary points in follow-up care. In cases where there is a low likelihood of malignancy, AI can be used to expedite such cases, thus relieving radiologists’ workload so they can focus on more complex cases.1,3

Listen above or on one of these podcast services:

iTunes
TuneIn
Stitcher
Spotify

References

1. McCrear S. AI-assisted mammogram readings reduce radiologist workload and maintain performance. AJMC®. August 20, 2025. Accessed October 22, 2025. https://www.ajmc.com/view/ai-assisted-mammogram-readings-reduce-radiologist-workload-maintain-performance

2. Taylor-Phillips S, Stinton C. Double reading in breast cancer screening: considerations for policy-making. Br J Radiol. 2020;93(1106):20190610. doi:10.1259/bjr.20190610

3. McCrear S, Adriana Olivo N. The future of survivorship care: precision tools for women after breast cancer. AJMC. September 10, 2025. Accessed October 22, 2025. https://www.ajmc.com/view/the-future-of-survivorship-care-precision-tools-for-women-after-breast-cancer

4. Screening for breast cancer. CDC. September 16, 2024. Accessed October 22, 2025. https://www.cdc.gov/breast-cancer/screening/index.html