New computer aided diagnostic system using deep neural network and SVM to detect breast cancer in thermography
Nabil Karim Chebbah, Mohamed Ouslim, Sorore Benabid
Abstract
Mammography is widely used for identifying breast cancer. However, this technique is invasive, which causes X-ray tissue damage and very often fails to detect a certain tumour size. Thermography is another alternative, being non-ionising, non-invasive and able to detect abnormal breast conditions at an early stage. In this paper, we propose a new computer-aided diagnosis system based on artificial intelligence and thermography to help radiologists correctly diagnose breast diseases. One hundred and seventy infrared breast images are collected from an open-source database to feed a deep learning algorithm for automatic segmentation of breast thermograms. An intersection over a union of 89.03% is practically obtained using the U-net model. Textural evaluation and vascular network analysis are performed on the segmented thermograms to extract relevant features. Classifiers based on supervised learning algorithms are implemented using the extracted features to distinguish normal from abnormal thermograms. . We achieved an accuracy of 94.4%, a precision of 96.2%, a recall of 86.7%, an F1-score of 91.2% and a true negative rate of 98.3% when the developed approach was applied on a support vector machine. These two obtained results concerning both segmentation and classification are considered very motivating and encouraging compared to up-to-date methods.