Multi-Disease Detection in Retinal Imaging Based on Ensembling Heterogeneous Deep Learning Models
Problem
Preventable or undiagnosed visual impairment and blindness affect billions worldwide, needing automated multi-disease detection in retinal imaging.
Why it matters
Undiagnosed retinal disease leads to preventable blindness, creating need for clinical decision support.
What the AI did
Built an ensemble pipeline combining several heterogeneous deep CNN models with transfer learning, class weighting, augmentation, Focal loss, 5-fold bagging, and stacked logistic regression for multi-disease retinal detection.
Results
Claims high accuracy, reliability, and comparability with state-of-the-art pipelines via internal and external evaluation; no numbers given
Impact
Evidence supports a validated ensemble pipeline, but no concrete metrics are stated.
Record
| Venue | Studies in health technology and informatics |
|---|---|
| Published | 2021 |
| Work type | book-chapter |
| Authors | Dominik Müller, Iñaki Soto‐Rey, Frank Krämer |
| Citations | 35 (OpenAlex, at import time) |
| References | 21 |
| Model or method | Heterogeneous deep CNN ensemble |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3233/shti210537 |
| Open access | hybrid |
| Keywords | retinal imaging, ensemble learning, deep CNN, multi-disease detection, transfer learning |
Importance rating
significant — thousands to millions affected
Abstract states preventable/undiagnosed visual impairment and blindness affect billions worldwide; automated multi-disease retinal detection offers clinical decision support at large scale.