Deep-CNN for Disease Classification Using Enhanced Mammographic Images

Purpose: Breast cancer has become one of the most common diseases that women face today as a result of poor nutrition and other environmental factors. A mammogram image of the breast will help detect breast cancer, but still, sometimes doctors and radiologists are unable to detect it due to poor image quality or abnormal region that appears to be normal. Materials and Methods: In this paper, a deep CNN-based classification model is proposed that classifies the mammogram image as normal, masses, and micro-calcification. Firstly, the PSNR values of the mammogram images is improved using a median filter with the Local Contrast Modification (LCM) method. It is further enhanced by Adaptive-CLAHE in con junction with the Wiener filter. After image enhancement, the region of interest is segmented through morphological feature extraction and the Otsu thresholding method. Results: In order to increase the number of samples in the mammogram image dataset, image data augmentation is applied to segmented images. Conclusion: Finally, a pre-trained ResNet model is used for the classification of mammogram images. The proposed model has shown improved PSNR for mammogram images and achieved a higher classification accuracy of 98.91%, thus outperforming other existing methods. Additionally, the explainability and causality of the proposed model are also discussed to show the learning process of the model.