EXPLAINABLE DEEP LEARNING FOR REDUCING FALSE NEGATIVES IN EARLY CANCER IMAGING DETECTION

Abstract

Early cancer detection remains a major clinical challenge because false-negative predictions can delay diagnosis, reduce treatment effectiveness, and increase patient risk. This study presents an explainable deep learning framework designed to reduce false negatives in medical imaging-based cancer detection while maintaining strong diagnostic reliability. The proposed model integrates convolutional feature extraction with attention-based interpretability to improve the identification of subtle malignant patterns that are often missed by conventional classifiers. Experimental evaluation was conducted using medical imaging data divided into training, validation, and testing subsets. The results demonstrated improved sensitivity, recall, and overall diagnostic performance compared with baseline deep learning models. In particular, the explainable model reduced false-negative cases by emphasizing clinically relevant image regions through Grad-CAM-based visual explanations. The model achieved high classification accuracy, stronger area under the ROC curve, and better calibration across multiple validation folds. The findings show that explainable deep learning can support early cancer screening by improving transparency, strengthening clinician trust, and reducing missed cancer detections. Overall, this research highlights the potential of interpretable artificial intelligence as a decision-support tool for safer, more reliable, and clinically useful cancer diagnosis from medical imaging data.