| Summary: | 碩士 === 國立高雄科技大學 === 電機工程系 === 107 === This paper uses an intelligent modeling method in macular degeneration identification. In this paper, a residual network (ResNet) is used to model different classifications of macular degeneration. For medical image processing application by using deep learning, many scholars only focused on the excellent accuracy for models, but they didn’t take the stability of the model accuracy after modeling into account. Therefore, this paper attaches important to the accuracy and standard deviation (SD) so that create a multi-objective intelligent modeling method.
In order to model effectively, this paper uses an experimental design method (uniform design, UD) to design an optimal model with two quality characteristics, the average and SD of the accuracy. Because the parameters of the model affect the overall efficiency, UD method is used to parameter selection in this paper. In UD method, the parameters need to adjust are defined firstly, and experimental layout is executed by using uniform layout. Then, the classified image data is put into Resnet and trained model with the experimental layout as well as the quality of the modeling is compared. Therefore, the models are expected to have better robustness and the model with the optimization parameter combination can be obtained, which have higher accuracy and lower SD. In this paper, in order to verify the effectiveness of the proposed method, the multi-objective S/N ratio of qualitative characteristics is used to calculate. Compared with model by using AlexNet, it is found that the S/N ratio of the best parameter combination in the uniform layout is 56.19 and it is higher than the result from AlexNet, which is 53.81. From the experimental result, the model by using ResNet with the best parameter combination from UD is better than AlexNet. Finally, the evaluation of the performance of ResNet and AlexNet is by using the test data. Compared with the model by using Najeeb et al. (2018), the model by the proposed method from this paper is excellent.
Therefore, the intelligent modeling method proposed by this paper can improve the doctor's confidence and sufficient diagnostic ability for the subsequent diagnosis of the macular image. It can also provide patients with maximum security and service satisfaction.
|
|---|
