Image Processing as an Auxiliary Methodology for Analysis of Thermograms

This paper presents a study by image processing to automate the thermogram analysis method of patients diagnosed with cancer. The objective is to develop a semiautomatic segmentation model of thermographic images using the Python computational language. A segmentation routine is proposed based on a region growth algorithm capable of grouping similar pixels to a Thermogram Region of Interest (ROI), starting from the manual positioning of the seed pixel, which is why the test is said to be semiautomatic. The tests were performed on twenty thermograms collected from patients with breast and thyroid cancer. As results it was verified that the proposed model comprises the tumor region with greater reliability than the manual delimitation method, thus the average and the minimum temperatures are higher (compared to the manual method) as it ensures that temperature points outside the real nodular range are not included in the ROI. As for operating time, the proposed method performs the ROI delimitation faster than the manual method. For future work, we suggest the statistical study for nodule benignity or malignancy based on thermal difference recorded in the ROI thermograms analyzed with the semiautomatic segmentation.