Deep learning-based M-mode OCT system and B-mode OCT system diagnosis accuracy comparison
Tong, Linjie
Loading…
Permalink
https://hdl.handle.net/2142/129191
Description
Title
Deep learning-based M-mode OCT system and B-mode OCT system diagnosis accuracy comparison
Author(s)
Tong, Linjie
Issue Date
2025-04-30
Director of Research (if dissertation) or Advisor (if thesis)
Do, Minh
Department of Study
Electrical & Computer Eng
Discipline
Electrical & Computer Engr
Degree Granting Institution
University of Illinois Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
Medical Image Analysis
Deep Learning
Language
eng
Abstract
M-mode Optical Coherence Tomography (OCT) imaging is a cost-effective alternative to the widely used B-mode OCT in medical imaging. However, as an emerging imaging modality, M-mode OCT has not been extensively studied for its diagnostic capabilities. In this paper, we propose a convolutional neural network (CNN)-based framework to evaluate the diagnostic performance of M-mode and B-mode OCT images. Our results demonstrate that M-mode OCT can achieve comparable diagnostic accuracy to B-mode OCT. To investigate the reason behind this comparable performance, we conduct further analysis in two parts. First, using transfer learning, we show that deep learning models extract highly similar features from both M-mode and B-mode OCT images. Second, we analyze the feature distributions and observe that both modalities yield distinguishable differences between normal and abnormal cases. These findings suggest that the critical diagnostic information in OCT images is primarily encoded in the depth profiles of individual A-scans. Motivated by this insight, we propose a weakly supervised algorithm based on Multi-Instance Learning (MIL), which extracts features from individual A-scans and integrates them to generate final diagnostic predictions. Notably, this method does not require per A-scan labels during training, yet it is capable of producing per A-scan predictions. The proposed approach achieves diagnostic performance comparable to models that utilize entire M-mode or B-mode OCT images, while offering enhanced interpretability through localized, per A-scan outputs.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
