New AI Model Diagnoses Breast Cancer with 98% Accuracy and Explains Its Reasoning

Early diagnosis of breast cancer is critical for improving survival rates, yet many existing artificial intelligence (AI) tools suffer from significant limitations. Current models often rely on a single data source and generally perform simple binary classification, determining only if cancer is present or not. Furthermore, they frequently operate as 'black boxes', offering a diagnosis without revealing the logic behind it, which limits their utility in a clinical setting.

To bridge this gap, researchers have proposed 'ResTab Net', a sophisticated model that integrates histopathological (tissue) images with molecular protein expression data. This multimodal approach allows for a more comprehensive analysis. The system employs specific algorithms to refine the input: Adaptive Tissue-Aware Gaussian Filtering enhances image quality, whilst Entropy Enhanced Graph-Watershed Segmentation precisely defines the tumour's location. A novel technique called Self-Adaptive Starfish Optimization is then used to select the most pertinent features for analysis.

Crucially, the researchers addressed the transparency issue to foster clinical trust. By integrating explainable AI tools—specifically SHAP and LIME—the model visualises exactly how molecular protein levels and visual features impact its classification results. This allows clinicians to see the 'why' behind the 'what'. Implemented in Python, the model achieved remarkable performance, boasting an accuracy of 98.56%, a precision of 98.10%, and an Area Under the Curve (AUC) of 99.60%, marking a significant step forward in reliable, automated diagnostics.