Breast cancer is the most commonly diagnosed form of cancer in women. Thermography is an efficient screening modality for detecting breast cancer as it is able to detect small tumors that can lead to earlier diagnosis. This study proposes a novel method for design an automatic system for breast thermal analysis. In this regard, main stages are database preparation, boundary detection, segmentation, abnormal pattern recognition and interpretation of the breast thermograms. First, we have provided an IUT-OPTIC database containing breast thermograms from three views, thermography report, mammography images , related report and patient information. Then we have presented a novel boundary detection algorithm using directional SUSAN algorithm with 98% accuracy. In the following, a novel extended hidden Markov model for segmentation of breast thermograms is presented that is able to map semi hot regions into distinct areas . In pattern recognition stage, we have presented a proof-of-concept scheme for classifying abnormal patterns using local histograms of vessel orientations and the error of classification is 0.118 on average. Finally, images are interpreted using features obtained in previous stages and random forest classifier with 14% error. We conclude that final presented automatic system has a remarkable impact on the analysis of breast thermal images. Key words
