| Summary: | 碩士 === 國立臺北科技大學 === 資訊與財金管理系 === 106 === Parotid gland tumors are one of the rare diseases of the human head and neck. There is no clear study of the cause and risk of the disease. Although the parotid gland tumors about 80% probability is benign tumors, but the tumor could gradually grew up and cause oppression adjacent normal tissues, even has the possibility of malignant change, therefore usually require surgical removal. In order to improve the efficiency of physician diagnosis, reduce human error and increase the accuracy of medical diagnosis, this study proposes a model to determine the parotid gland tumor. The model could help physicians to found the exact location of the tumor. The discriminant model proposed in this study contains two phases, namely, "data entry" and "discriminant mechanism". In the first, this study was based on the magnetic resonance imaging provided by the Radiation Diagnostic Department of a medical center in Taipei to training and adjustment the parameters of the tumor condition, and using Convolutional Neural Networks to train and tune the models through deep learning. Then, after the training is completed, the accuracy of the new image test system is determined by the training and identification system.
In this study, the accuracy rate of actual test results was more then 99%. The experimental results showed that this model can successfully predict the patient is suffering from parotid gland tumors. The relevant rules of the predictive results can be applied to the postoperative examination of the parotid gland tumor. Physician can detect if patient had tumor and distinguish the actually position of the tumor. Finally, assisting physicians make more accurate diagnosis.
|
|---|
