Off-campus SJSU users: To download campus access theses, please use the following link to log into our proxy server with your SJSU library user name and PIN.
Publication Date
Fall 2025
Degree Type
Thesis - Campus Access Only
Degree Name
Master of Science (MS)
Department
Computer Engineering
Advisor
Taehee Jeong; Carlos Rojas; Haonan Wang
Abstract
Diabetic Retinopathy (DR) lesion segmentation is critical for early detection but challenged by extreme lesion scale variation, from tiny microaneurysms (MAs) to large hemorrhages (HEs). This thesis investigates the impact of input resolution scaling on different lesion types. Through systematic experimentation with multiple architectures (U-Net, UNet++, DeepLabV3+) at 512×512 and 1024×1024 resolutions, we identify a critical, counter-intuitive phenomenon where increasing input resolution has opposing effects on different lesion types. We demonstrate that while higher resolution is essential for resolving fine-grained MAs, it can simultaneously and unexpectedly degrade performance on larger HEs, with certain architectures exhibiting a severe trade-off. This finding challenges the common assumption that higher resolution is uniformly beneficial. To address this, we propose a novel Multi-Resolution Feature Stem, an input-level pyramid integrated with a UNet++ backbone. This architecture processes multiple scales in parallel, capturing fine-grained details without sacrificing contextual information. This work contributes crucial empirical evidence of this complex, resolution-dependent behavior and a practical, parameter-efficient architecture that successfully resolves this trade-off.
Recommended Citation
Dutta, Indranil, "An Input-Level Multi-Resolution Stem for UNet++ in Diabetic Retinopathy Lesion Segmentation" (2025). Master's Theses. 5730.
DOI: https://doi.org/10.31979/etd.ujmv-j4km
https://scholarworks.sjsu.edu/etd_theses/5730
