Deep Retinal Fundus Image-Based Alzheimer’s Disease Diagnosis Network for Mobile Devices
모바일 기기용 안저 영상 기반 알츠하이머병 진단 딥러닝 모델
- 주제(키워드) ai diagnosis , alzheimer's disease , deep learning
- 주제(DDC) 621.381
- 발행기관 아주대학교
- 지도교수 선우명훈
- 발행년도 2023
- 학위수여년월 2023. 2
- 학위명 석사
- 학과 및 전공 일반대학원 전자공학과
- 실제URI http://www.dcollection.net/handler/ajou/000000032412
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
The leading cause of dementia is Alzheimer's disease (AD), a neurodegenerative disorder that is still difficult to identify in its early stages. Numerous deep-learning methods have recently been introduced for diagnosing AD. However, most current approaches use magnetic resonance (MR), computed tomography (CT), and positron emission tomography (PET) images as datasets, which demand a significant amount of processing resources for deep learning networks to make calculations based on these medical scans. To improve AD diagnosis accessibility, the goal of this study is to present a novel lightweight mobile device-friendly deep learning network model for detecting Alzheimer's disease using retinal fundus images. MobileNetV3 is used as the backbone in the proposed model, with a structure change based on the UNet design. For better diagnosis outcomes, we used an attention mechanism and transformed it into a weighted scaled dot-product attention mechanism. Training with image masking was used to improve the model's robustness. By reaching a 0.927 area under the receiver operating characteristic curve (AUC) on the validation dataset from the UK Biobank, our model demonstrated strong AD diagnostic performance. The experimental results indicated that the proposed method, which has 4.42 M parameters, is mobile device-friendly. Additionally, we successfully emulated the suggested approach and tested how it could operate in a mobile application. Medical diagnostic techniques based on deep learning are developing quickly. However, the majority of the existing AD diagnosis techniques take a lot of time and money. We can verify that the suggested approach can improve diagnostic performance while using fewer of these resources. This study shows how the proposed method could be applied to cell phones as a clinical AD clinical tool. The presented model may be used as a self-test tool for regular people in addition to helping professionals detect AD with more validation.
more목차
I. Introduction 1
II. RELATED WORKSRELATED WORKS 3
A. CT-based method 3
B. MRI-based methods 3
C. PET-based methods 5
D. Multimodal method 5
E. Cerebral angiography-based method 6
F. AD classification based on the retinal fundus vasculature 6
III. MATERIALS AND METHODS 7
A. Dataset description 7
B. Proposed model 8
C. Experimental method 14
IV. Experimental Results 15
A. Diagnosis performance comparison 16
B. Image quality robustness 18
C. Implementation 19
Bibliography 22
