Historically, the skin temperature has been used as a functional indicator of the processes undergoing inside the body, where the anomalous change in temperature gives indication of illness. Infrared thermography (IRT) is a non-invasive approach which radiometrically measures the temperature distribution on the object’s surface based on emanating IR radiation. In medical field, IRT has been used to diagnose the various problems associated with superficial body parts since 1960s. The clinical abnormalities can be diagnosed by visual and subjective analysis of the thermograms, but human based diagnoses are likely to be influenced by narcissus effect, negligence, visual exhaustion and mental workload. Apart, the deep and objective diagnostic information, like severity and type of a disease can only be obtained by statistically analyse the region of interest (ROI). So, the segmentation of ROI has often been an initial step in medical diagnosis, where the automated approaches lead to fast and highly reproducible analysis. Recent advancement in deep learning opens a new way for effective computer-aided medical diagnosis, several studies have proposed deep learning based automatic segmentation and data classification. This review focuses on the wide range of image processing techniques adopted for automatic segmentation of various clinically significant ROIs, like breast, face, foot, hand etc. In addition, the present efforts are targeted on the selection of characteristic points for an effective automatic segmentation and identification of new areas to carry out the further research. Besides, the basics of medical IRT, need of automatic segmentation and factors which hinder the automatic segmentation in IRT are also discussed.
Read full abstract