MMSHRs: a morphology model of suspicious hyperthermic regions for degree of severity prediction from breast thermograms

ABSTRACT The presence of suspicious hyperthermic regions (SHRs) in breast thermograms is a prominent indicator of breast pathology, for which delineation and analysis of SHRs have a crucial role in early detection of breast abnormalities. A novel approach for breast abnormality grading, namely the morphology model of suspicious hyperthermic regions (MMSHRs), is proposed here. The proposed model first segments SHRs from breast-thermograms and then analyzes their morphology to grade the thermograms according to their degree of severity. To segment SHRs, a simple but effective method that computes the similarity score of each pixel with the highest intensity value is designed. . The performance of the proposed segmentation method is tested on both public and in-house-captured datasets. With the optimal values of seven evaluation metrics, the proposed segmentation method outperforms other state-of-the-art segmentation methods. The values of evaluation metrics further justify that the proposed SHRs segmentation method addresses all the limitations regarding infrared breast thermogram segmentation, and reduces the under-segmentation and over-segmentation of SHRs. Following segmentation of SHRs, the MMSHRs extract the corresponding morphological features, allowing the classification of thermograms into mild and severely abnormal with the classification accuracy of 91% and area under the receiver operating characteristic curve of .9998.