| Summary: | 碩士 === 國立臺灣大學 === 資訊網路與多媒體研究所 === 100 === Breast cancer is a common disease and a major cause of death for women. Cancer patients often receive chemotherapy to stop the proliferation of cancer cells, but the side effects of chemotherapy also injure them greatly. The early prediction of therapy results not only helps to decide whether the change of treatment is needed to prevent damage that unnecessary therapy course brings to the patients, but also is useful in evaluating the need of surgical treatment. In general, the therapy result is predicted by the residual tumor size, yet the tumor vascularity is also a potential prognostic indicator. Research in the past has broadly used the vascularization index to evaluate the response of chemotherapy. But the lack of morphological and functional description of this indicator makes the experiment results have lower reliability. In this paper, we used 3-D power Doppler to observe vessel information. Due to the departure of the scanned region, the image registration technique is applied to align the vessels. The vessels in the intersected region will be extracted and skeletonized to facilitate the quantification of vessel features. To complement the insufficiency of morphological and functional description, seventeen vascular features depicting changes in morphology and tortuosity are traced and the change ratio will be used to predict the therapy response. The proposed system was examined on 32 breast cancer patients that underwent neoadjuvant chemotherapy, including 10 negative-response and 22 positive-response cases. The accuracy, sensitivity, specificity, positive and negative predictive values were 84.4 (27/32), 86.4 (19/22), 80.0 (8/10), 90.5 (19/21) and 72.2 (8/11), respectively. The Az value of the ROC curve was 0.91. The results showed a high correlation between the change of vessels and response to chemotherapy and the proposed system can differentiate the therapy results into negative or positive responses.
|
|---|
