Answers are grounded in this paper and cite their sources.
Breast cancer is the leading malignant disease in women globally, with 2.3 million new diagnoses and 685,000 deaths recorded annually. Clinical deployment of AI-based diagnostic tools has been impeded by model opacity, miscalibrated confidence, and unequal performance across patient subgroups. Methods: We propose TEDIN (Trustworthy Explainable Deep Intelligence Network), a multi-module framework that couples a ResNet-50 transfer learning backbone with Gradient-weighted Class Activation Mapping (Grad-CAM) for spatial explainability, Monte Carlo (MC) Dropout for epistemic uncertainty quantification, and equalized-odds post-processing for demographic fairness correction. TEDIN is evaluated on the CBIS-DDSM mammography dataset (2568 images) and BreakHis histopathology dataset (7909 images). Results: TEDIN achieves 97.4% accuracy, 96.8% sensitivity, 97.9% specificity, and AUC = 0.989, outperforming four state-of-the-art baselines. The MC-Dropout mechanism flags 88.2% of misclassified cases as high-uncertainty. Equalized-odds correction reduces cross-demographic false-negative rate disparities by up to 74.5% at less than 0.4 pp accuracy cost. A usability study with 12 radiologists confirms a 23% reduction in diagnostic review time and statistically significant gains in clinician confidence (p < 0.01). Conclusions: TEDIN demonstrates that the reliability, interpretability, calibration, and fairness principles central to Trustworthy and Intelligent Systems for Predictive Maintenance transfer directly and productively into clinical breast cancer diagnosis.
No takes yet. Share an insight, caveat, or question.
Padmaja et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: