A novel explainable AI model accurately detects and localizes breast tumors on MRI, outperforming conventional models—especially in low-cancer-prevalence screening scenarios.
Key Details
- 1The explainable fully convolutional data description (FCDD) model was trained and tested on 9,738 breast MRI exams from 2005-2022, plus an external multicenter dataset.
- 2FCCD outperformed standard binary classification models, achieving AUCs up to 0.84 (balanced tasks) and 0.72 (imbalanced) vs. 0.81 and 0.69 for benchmarks (p<0.001).
- 3In internal and external validation, FCCD consistently showed higher detection performance, e.g., AUC of 0.86 vs. 0.79 (external set).
- 4The model achieved a specificity of 13% vs. 9% for the benchmark at 97% sensitivity in imbalanced (realistic) settings (p=0.02).
- 5It produces interpretable heatmaps to highlight probable tumor areas, addressing the 'black box' issue in AI models.
- 6Researchers note potential to streamline breast MRI screening, including use with abbreviated MRI protocols.
Why It Matters
Demonstrating robust detection of breast cancer in both high and low prevalence MRI datasets, this explainable AI system advances the field toward more transparent, generalizable tools for radiologists. Its validated, interpretable outputs may enhance clinical adoption and trust in AI-based breast imaging solutions.
Source
AuntMinnie
Related News
•AuntMinnie
Multimodal LLMs Show Improved Performance on Japanese Radiology Board Exams
New multimodal large language models (LLMs) like OpenAI o3 and Gemini 2.5 Pro demonstrated significant advancements in answering Japanese radiology board exam questions, particularly with image input.
•Health Imaging
AI Surpasses Radiologists in Predicting Lung Cancer Treatment Response
AI demonstrates higher accuracy than radiologists in predicting lung cancer treatment response from imaging.
•Radiology Business
Medicare Denies Over Half of Radiology AI Claims, Study Finds
Medicare denied $16 million in radiology AI claims between 2018 and 2023, rejecting over half submitted.