Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Balancing Accuracy and Efficiency: CNN Fusion Models for Diabetic Retinopathy Screening

Md Rafid Islam, Rafsan Jany, Akib Ahmed, Mohammad Ashrafuzzaman Khan

TL;DR

The paper tackles scalable diabetic retinopathy screening by balancing accuracy and throughput across heterogeneous fundus images. It investigates feature-level fusion of complementary CNN backbones (ResNet50, EfficientNet-B0, DenseNet121) and evaluates pairwise and tri-fusion strategies on a pooled dataset of 11,156 images from five public sources. The EfficientNet-B0 + DenseNet121 fusion achieves the best mean accuracy (82.89%) with balanced class-wise F1-scores, while tri-fusion offers higher recall at a substantial computational cost. The results support lightweight fusion as a practical approach for robust, scalable DR screening in resource-limited settings, with deployment options depending on latency constraints and throughput needs.

Abstract

Diabetic retinopathy (DR) remains a leading cause of preventable blindness, yet large-scale screening is constrained by limited specialist availability and variable image quality across devices and populations. This work investigates whether feature-level fusion of complementary convolutional neural network (CNN) backbones can deliver accurate and efficient binary DR screening on globally sourced fundus images. Using 11,156 images pooled from five public datasets (APTOS, EyePACS, IDRiD, Messidor, and ODIR), we frame DR detection as a binary classification task and compare three pretrained models (ResNet50, EfficientNet-B0, and DenseNet121) against pairwise and tri-fusion variants. Across five independent runs, fusion consistently outperforms single backbones. The EfficientNet-B0 + DenseNet121 (Eff+Den) fusion model achieves the best overall mean performance (accuracy: 82.89\%) with balanced class-wise F1-scores for normal (83.60\%) and diabetic (82.60\%) cases. While the tri-fusion is competitive, it incurs a substantially higher computational cost. Inference profiling highlights a practical trade-off: EfficientNet-B0 is the fastest (approximately 1.16 ms/image at batch size 1000), whereas the Eff+Den fusion offers a favorable accuracy--latency balance. These findings indicate that lightweight feature fusion can enhance generalization across heterogeneous datasets, supporting scalable binary DR screening workflows where both accuracy and throughput are critical.

Balancing Accuracy and Efficiency: CNN Fusion Models for Diabetic Retinopathy Screening

TL;DR

The paper tackles scalable diabetic retinopathy screening by balancing accuracy and throughput across heterogeneous fundus images. It investigates feature-level fusion of complementary CNN backbones (ResNet50, EfficientNet-B0, DenseNet121) and evaluates pairwise and tri-fusion strategies on a pooled dataset of 11,156 images from five public sources. The EfficientNet-B0 + DenseNet121 fusion achieves the best mean accuracy (82.89%) with balanced class-wise F1-scores, while tri-fusion offers higher recall at a substantial computational cost. The results support lightweight fusion as a practical approach for robust, scalable DR screening in resource-limited settings, with deployment options depending on latency constraints and throughput needs.

Abstract

Diabetic retinopathy (DR) remains a leading cause of preventable blindness, yet large-scale screening is constrained by limited specialist availability and variable image quality across devices and populations. This work investigates whether feature-level fusion of complementary convolutional neural network (CNN) backbones can deliver accurate and efficient binary DR screening on globally sourced fundus images. Using 11,156 images pooled from five public datasets (APTOS, EyePACS, IDRiD, Messidor, and ODIR), we frame DR detection as a binary classification task and compare three pretrained models (ResNet50, EfficientNet-B0, and DenseNet121) against pairwise and tri-fusion variants. Across five independent runs, fusion consistently outperforms single backbones. The EfficientNet-B0 + DenseNet121 (Eff+Den) fusion model achieves the best overall mean performance (accuracy: 82.89\%) with balanced class-wise F1-scores for normal (83.60\%) and diabetic (82.60\%) cases. While the tri-fusion is competitive, it incurs a substantially higher computational cost. Inference profiling highlights a practical trade-off: EfficientNet-B0 is the fastest (approximately 1.16 ms/image at batch size 1000), whereas the Eff+Den fusion offers a favorable accuracy--latency balance. These findings indicate that lightweight feature fusion can enhance generalization across heterogeneous datasets, supporting scalable binary DR screening workflows where both accuracy and throughput are critical.
Paper Structure (24 sections, 11 equations, 5 figures, 3 tables)

This paper contains 24 sections, 11 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Dataset Description
  5. Data Preprocessing and Augmentation
  6. Base Models
  7. ResNet50
  8. EfficientNetB0
  9. DenseNet121
  10. Fusion Model
  11. Fusion Strategy
  12. Fusion Architecture
  13. Training Procedure
  14. Evaluation Metrics and Testing
  15. Accuracy
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Comprehensive DR detection workflow: preprocessing, training, and evaluation.
  • Figure 2: Architecture of the best-performing pairwise fusion model (EfficientNet-B0 + DenseNet121).
  • Figure 3: Confusion matrices for all models from the best-performing run.
  • Figure 4: Comparative visualization of evaluation metrics across all models.
  • Figure 5: Inference latency across batch sizes for base models and the Eff+Den fusion model: (a) total batch inference time (seconds); (b) per-image inference time (milliseconds).