A Methodology for Breast Disease Computer-Aided Diagnosis using dynamic thermography

Breast cancer is the one that most affects women worldwide. Thus, disease screening techniques are constantly being developed to detect the disease at early stages. The dynamic thermography image analysis emerges as one tool to aid in the diagnosis of breast diseases. Thermal images captured by the dynamic protocol were used in this thesis where, after the cooling of the breasts by air stream, 20 sequential images with intervals of 15 seconds between them were taken during the process of returning the patient’s body to thermal equilibrium with the enviroment. The images are stored in the Database for Research Mastology with Infrared Image - DMR-IR, accessible on the website http://visual.ic.uff.br/dmi. After obtaining the sequential thermal images of each patient, their temperature matrices were extracted and new grayscale images were generated. Then the regions of interest (ROIs) of the images in grayscale were segmented and from them also the ROIs of the temperature matrices were determined. Features were extracted from the ROIs (they are based on statistical, on clustering, on histogram comparison, on fractal geometry, on diversity indices and on spatial statistics). Time series which were broken down into 9 subsets of different cardinalities were generated from these features. The best features were selected and classified by the Support Vector Machine (SVM), using the Leave-One-Out Cross-Validation method. An accuracy of 100% was obtained for a sample of 64 breast (composed of 32 healthy and 32 with some abnormality). Tests were performed with 12 new breasts that did not participate in the methodology development stage. The method also achieved 100% accuracy with them. At the end a methodology of computer-aided diagnosis (CAD) system was proposed based on sequential thermography.


Download image database

Camera used by the project: FLIR SC-620,
Resolution: 640 x 480 pixels