| dc.description.abstract | The kidneys are vital organs in the human body. When kidney function declines to a certain level, renal replacement therapy, such as dialysis, becomes necessary. Unfortunately, the decline of kidney function can lead to vascular calcification, causing blood vessels to lose their elasticity and ultimately increasing mortality risk. Currently, the assessment of abdominal aortic calcification is performed by specialized physicians. However, this approach has limitations as the scoring is subjective, resulting in inconsistent ratings between different physicians or even the same physician at different times. Occasionally, there may also be erroneous segment scoring of the abdominal aorta.
This study aims to find a method to help physicians assess the degree of vascular calcification in patients more objectively. We utilized models such as Logistic Regression, Ridge, SVM_Linear, K Nearest Neighbor, Random Forest, and MLP for physiological data. For X-ray images, we employed models including Resnet50 with and without pretraining and Vision Transformer. Currently, the best-performing model for physiological data is the Logistic Regression model trained using RidgeCV to select the top three important factors, achieving an accuracy of 67.22%. Regarding image analysis, the Vision Transformer model achieved the highest accuracy of 54.17%. However, when training the Teachable Machine model using the original images and images with background masking separately, the accuracies reached 75% and 81.25%, respectively. This indicates that masking the unobserved regions as background effectively avoids interference in the model’s judgment caused by the background. | en_US |