Explainability-Based Optimized Deep Learning in Histopathological Diagnosis of Multiple Cancers and Development of...

Area of Science:

• Medical image analysis
• Computational pathology
• Artificial intelligence in oncology

Background:

• Histopathological image analysis is crucial for cancer diagnosis but faces challenges in interpretability, computational cost, and accuracy.
• Existing models often lack transparency, hindering clinical trust and adoption.

Purpose of the Study:

• To introduce a novel deep learning model, Complementary Residual Retentive Network with Guided Gaussian Combined Arms Algorithm (C2RN2GC2A), for enhanced cancer classification in histopathological images.
• To improve model efficiency, accuracy, and interpretability in diagnostic tasks.

Main Methods:

• Developed C2RN2GC2A, integrating residual learning with Gaussian perturbations for feature extraction and efficiency.
• Employed the Two-stage Guided Chaotic Capuchin Algorithm (2GC2A) for metaheuristic optimization, feature selection, and accelerated convergence.
• Utilized Layer-wise DeepLIFT Relevance Propagation (LDLRP) for explainable AI, identifying critical image regions influencing classification.

Main Results:

• Achieved 98.02% accuracy on the LC25000 dataset with 0.08 training loss.
• Attained 98.54% accuracy on the BreakHis dataset with 0.05 validation loss.
• Demonstrated significant improvements in diagnostic reliability and clinical transparency.

Conclusions:

• The C2RN2GC2A framework offers a robust solution for accurate and interpretable histopathological image analysis.
• The integration of advanced AI techniques enhances diagnostic capabilities and clinical trust in cancer classification.