FINE-TUNING A MODIFIED XCEPTION ARCHITECTURE FOR ENHANCED BREAST CANCER DETECTION IN MAMMOGRAMS: AN ANALYTICAL STUDY

This research focuses on the adaptation and optimization of the Xception deep learning model for the detection of breast cancer using mammogram images, contributing to the field of precision medicine. Breast cancer, a leading cause of morbidity globally, benefits significantly from early and precise detection. Our study leverages transfer learning to modify the Xception architecture, introducing custom adjustments to its final layers to better capture the subtle signs of malignancy in mammograms. These modifications are aimed at improving the model's accuracy without compromising its specificity. The model was optimized using the Adam optimizer alongside the ReLU activation function, which helped in dynamic learning rate adjustment and enhanced feature detection from complex mammographic images. We assessed the model's performance through a comprehensive set of metrics including accuracy (90.42%), precision, recall, F1 score, ROC AUC score (0.9388), and Cohen's Kappa. These metrics collectively suggest a robust improvement over traditional detection methods, with a significant reduction in false positives and negatives. This research findings show that the modified Xception model not only meets but exceeds expectations in terms of diagnostic accuracy and reliability, offering a potent tool for clinicians. This research not only advances the application of deep learning in medical imaging but also paves the way for further research into AI-assisted diagnosis, potentially transforming clinical decision-making processes in breast cancer management. The implications of this work extend beyond immediate diagnostic benefits, supporting a broader shift towards more personalized and effective healthcare solutions.