An affordable solution for the recognition of abnormality in breast thermogram

Abstract

Lack of sufficient expertise in the rural regions of the country contributes to a higher mortality rate of breast cancer. Remote breast health monitoring systems, including image acquisition devices and advanced communication technologies, have been laid out of a new lease of life by the conveyance of quality healthcare services in developing parts of the world. Despite the high mortality rate of breast cancer, very limited existing works have been explored in integrating screening techniques with machine learning approaches and real-time communication to remote areas, secondary or tertiary hospitals. This approach is necessary to develop scalable and affordable breast screening technologies for clinical prediction of breast abnormality in the remote regions of the country. In this research work, we propose an affordable and portable infrared imaging solution for remote breast health monitoring. The proposed system integrates an Infrared Image Acquisition Module (IIAM), Screening Module (SM), and Transmission Module (TM). The IIAM includes a thermal camera and associated software to acquire thermal images of the breast. SM is the combination of four submodules such as Pre-processing Module (PM), Automatic Segmentation Module (ASM), Feature Extraction Module (FEM), and Classification Module (CM). The key challenge in implementing SM is that the penetration of thermography based diagnostic approaches are impeded by the frequent misclassifications in the diagnosis of breast cancer. The main reasons for this misclassification is the poor Signal to Noise Ratio (SNR) and inefficient segmentation of breast regions in thermograms. To address these challenges, co-occurrence filter-based edge-preserved technique is adopted to design the PM. Using morphological operations and Distance Regularized Level Set Evolution (DRLSE), ASM delineates the Region of Interest (ROI). FEM extracts both statistical features, and wavelet transform based features from the segmented breast ROI’s. CM depends on the SVM classifier to predict normal and abnormal images in the compiled dataset. The TM accesses and transmits the breast thermograms, predicted results, and patient’s history to the healthcare professionals in the tertiary hospitals for further diagnosis. Detailed in-person screening and experimentation was performed on 71 patients which consisted of 34 healthy and 37 abnormal images. The performance of the proposed solution is evaluated, which demonstrated a classification accuracy of 96.46% competitive compared to state-of-the-art schemes.