Deep learning based technique for classification of abdominal aortic aneurysm (AAA) in ct-scan images
1
Issued Date
2023
Copyright Date
2023
Resource Type
Language
eng
File Type
application/pdf
No. of Pages/File Size
xi, 56 leaves : ill.
Access Rights
open access
Rights
ผลงานนี้เป็นลิขสิทธิ์ของมหาวิทยาลัยมหิดล ขอสงวนไว้สำหรับเพื่อการศึกษาเท่านั้น ต้องอ้างอิงแหล่งที่มา ห้ามดัดแปลงเนื้อหา และห้ามนำไปใช้เพื่อการค้า
Rights Holder(s)
Mahidol University
Bibliographic Citation
Thesis (M.Sc. (Computer Science))--Mahidol University, 2023
Suggested Citation
Miao, Zhiqi, 1998- Deep learning based technique for classification of abdominal aortic aneurysm (AAA) in ct-scan images. Thesis (M.Sc. (Computer Science))--Mahidol University, 2023. Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/115336
Title
Deep learning based technique for classification of abdominal aortic aneurysm (AAA) in ct-scan images
Author(s)
Abstract
Abdominal aortic aneurysm (AAA) is known as an extremely dangerous cardiovascular disease for the elderly, with a high probability of death within an exceedingly short period of time. Therefore, the goal of our proposed method is to create a model that can assist radiologists in automatically and precisely determining the presence of AAA in the patient’s body, both during the patient’s daily physical examination and as soon as the patient has symptoms of AAA rupture. In our work, three state-of-the-art deep convolutional neural networks (ResNet, DenseNet, and EfficientNet) are adopted for 2D-based AAA classification on the non-contrast and contrast CT datasets. Based on our final patient-based fusion data division of training, our AAA-oriented model outperforms other existing works or has similar performance on classifying AAA (97% accuracy, 89% precision, 93% sensitivity, 98% specificity, 91% F1-score, and 95% ROC-AUC) tested by CT, CTA, and CT&CTA slices. Our work can also provide correspondingly accurate CAM visualization as a reference for AAA determination, both CT and CTA as inputs, and greater ease of use. Overall, our work could bring much clinical value to the radiologists in assisting the precise diagnosis and benefit the health of patients by reducing the probability of compulsory use of CTA with health-harmful contrast agents for the precise diagnosis. Implication of the thesis: (i) The precise and quick determination of the AAA with the heatmap as a reference of our work would save a great deal of time and workload for the radiologists. (ii) Instead of using only CTA with dangerous contrast agent injection in most existing works, both CT and CTA slices as the input could generate the accurate classification of the AAA in our work. (iii) Both CT and CTA slices could be the input for the accurate classification. (iv) The proposed work would be helpful to save more people from this life-threatening cardiovascular disease before the rupture.
Degree Name
Master of Science
Degree Level
Master's degree
Degree Department
Faculty of Information and Communication Technology
Degree Discipline
Computer Science
Degree Grantor(s)
Mahidol University