Automatic Detection and Classification of Thyroid Nodules by Thermal Imaging

Thermography is one of the non-invasive techniques that uniquely identify affected parts of the body. It senses temperature variation between diseased and normal area even at much earlier stage of the disease. Thyroid disease is a problem that affects other important organs of the body. Abnormal blood flow in the affected thyroid glands result in rise of skin temperature. This temperature variation is captured by the thermal camera. Analysis of thermal images is under research development for efficient disease detection and health management. The real-time thermal images captured by the FLIR thermal camera is subjected to pre-processing for possible reduction of noises. Pre-processing of the original thermal image is obtained using discrete wavelet transform. After noise removal, region of interest is segmented and the desired features are extracted by Gray level co-occurrence matrix (GLCM). Features of abnormal and normal thyroid glands are classified using multi-class Support vector machine (SVM) classifier. Performance evaluation metrics for the three classes were obtained respectively as, Normal: Precision- 99.01%, recall- 96.5%, F1 score- 97.8%; Hyper-thyroid: Precision- 89.8%, recall- 89.2%, F1 score- 89.8%; Hypo-thyroid: Precision- 96.7%, recall- 97.4%, F1 score- 96.5%. The results haven shown better classification accuracy of 98.07% as an average with three classes such as hyperthyroid, hypothyroid and normal thermal dataset.