A Novel Method for Medical Image Classification

Subject Area: Image Processing

Breast cancer stands out as one of the most noticeable and impassive diseases worldwide, primarily affecting women. Detecting it at its initial stage is crucial for an effective management. Many cases have shown diminish mortality rates through early identification. Extensive research is sustained in this field, with deep learning being a commonly utilized method. This study presents an efficient breast cancer classification model, employing weight-modified ResNet14 (WMRESNET) to extract features more accurately and rapidly, even with a small training dataset. The proposed model incorporates SVM classifier, that classifies different types of breast cancers based on the feature extracted by WMRESNET achieving a remarkable classification efficiency of 96% on Ultrasound Breast Cancer Images (UBSI) dataset. Pre-processing techniques such as Histogram Equalization and CNN-based denoising enhanced the feature extraction capability of WMRESNET during training. The objective of this research is to classify cancer images as benign, malignant, or non-cancerous, crucial for early detection and appropriate treatment. Comparative analysis with state-of-the-art classifier models demonstrates the superiority of the proposed model in terms of computational cost and classification accuracy.