| Summary: | 碩士 === 國立清華大學 === 電機工程學系所 === 106 === The mortality rate of breast cancer among women in Taiwan ranks the top five in 2017 according to the statistics of the cause of death of the Ministry of Health and Welfare, Taiwan. Hence early screening/diagnosis via mammogram or ultrasound of breast cancer is vital. Due to the painful process of taking mammogram and fear of excess X-ray radiation exposure, ultrasound breast image examination become cheaper and well-accepted in breast diagnosis. Tumor segmentation of ultrasound breast (USB) images becomes important.
Usually, traditional image segmentation methods are applied to obtain tumor segmentation of USB images. In this thesis, an AI-deep learning technique, namely the Mask-RCNN is adopted to solve the USB tumor segmentation problem. In addition to the speed advantage of the Faster-RCNN, the Mask-RCNN can also provide instance segmentation, which is suitable for tumor segmentation of USB images.
Several error measures (ME, RFAE, MHD, TPR, TNR, ACC) are used to evaluate the segmentation results of this thesis. Comparisons with the other four methods (retracting DRLSE, expansion DRLSE, the algorithm proposed by Tsai-Wen Niu, and the algorithm proposed by Hsun Hsieh) are made. It is found that the segmentation result of this thesis is superior to the other four methods in terms of ME, RFAE, MHD.
|
|---|
