Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Novel Fundus Image Preprocessing for Retcam Images to Improve Deep Learning Classification of Retinopathy of Prematurity

Sajid Rahim, Kourosh Sabri, Anna Ells, Alan Wassyng, Mark Lawford, Linyang Chu, Wenbo He

TL;DR

This work tackles automated ROP screening by addressing Retcam image quality via novel fundus preprocessing that complexes PCA-based illumination enhancement and DPFR-based restoration, augmented with CLAHE and vessel-erosion via segmentation. By pairing these preprocessing pipelines with transfer-learning CNNs (ResNet50, InceptionResv2, Dense169), the study achieves high accuracy for Plus disease and notable improvements for Stage (0–3) and Zone (I–III) classifications on a publicly described Calgary dataset. The results demonstrate that restoring and enhancing salient retinal features before CNN classification yields competitive, and often superior, performance to prior ROP deep-learning approaches, highlighting potential benefits for telemedicine deployment and broader fundus imaging. Limitations include small sample size and lack of independent testing, but the work outlines clear pathways for open datasets, segmentation-driven Zone analysis, and multi-task classifiers that could translate into improved remote screening workflows.

Abstract

Retinopathy of Prematurity (ROP) is a potentially blinding eye disorder because of damage to the eye's retina which can affect babies born prematurely. Screening of ROP is essential for early detection and treatment. This is a laborious and manual process which requires trained physician performing dilated ophthalmological examination which can be subjective resulting in lower diagnosis success for clinically significant disease. Automated diagnostic methods can assist ophthalmologists increase diagnosis accuracy using deep learning. Several research groups have highlighted various approaches. Captured ROP Retcam images suffer from poor quality. This paper proposes the use of improved novel fundus preprocessing methods using pretrained transfer learning frameworks to create hybrid models to give higher diagnosis accuracy. Once trained and validated, the evaluations showed that these novel methods in comparison to traditional imaging processing contribute to better and in many aspects higher accuracy in classifying Plus disease, Stages of ROP and Zones in comparison to peer papers.

Novel Fundus Image Preprocessing for Retcam Images to Improve Deep Learning Classification of Retinopathy of Prematurity

TL;DR

This work tackles automated ROP screening by addressing Retcam image quality via novel fundus preprocessing that complexes PCA-based illumination enhancement and DPFR-based restoration, augmented with CLAHE and vessel-erosion via segmentation. By pairing these preprocessing pipelines with transfer-learning CNNs (ResNet50, InceptionResv2, Dense169), the study achieves high accuracy for Plus disease and notable improvements for Stage (0–3) and Zone (I–III) classifications on a publicly described Calgary dataset. The results demonstrate that restoring and enhancing salient retinal features before CNN classification yields competitive, and often superior, performance to prior ROP deep-learning approaches, highlighting potential benefits for telemedicine deployment and broader fundus imaging. Limitations include small sample size and lack of independent testing, but the work outlines clear pathways for open datasets, segmentation-driven Zone analysis, and multi-task classifiers that could translate into improved remote screening workflows.

Abstract

Retinopathy of Prematurity (ROP) is a potentially blinding eye disorder because of damage to the eye's retina which can affect babies born prematurely. Screening of ROP is essential for early detection and treatment. This is a laborious and manual process which requires trained physician performing dilated ophthalmological examination which can be subjective resulting in lower diagnosis success for clinically significant disease. Automated diagnostic methods can assist ophthalmologists increase diagnosis accuracy using deep learning. Several research groups have highlighted various approaches. Captured ROP Retcam images suffer from poor quality. This paper proposes the use of improved novel fundus preprocessing methods using pretrained transfer learning frameworks to create hybrid models to give higher diagnosis accuracy. Once trained and validated, the evaluations showed that these novel methods in comparison to traditional imaging processing contribute to better and in many aspects higher accuracy in classifying Plus disease, Stages of ROP and Zones in comparison to peer papers.
Paper Structure (24 sections, 3 equations, 5 figures, 18 tables, 6 algorithms)

This paper contains 24 sections, 3 equations, 5 figures, 18 tables, 6 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. ROP detection based CNNs
  4. Fundus Pre-processing
  5. Image domain methods
  6. Restoration methods
  7. Methods
  8. Background
  9. Datasets
  10. Traditional image pre-processing methods
  11. Pixel Colour Amplification for ROP Retcam
  12. Double Pass Fundus Reflection for ROP Retcam
  13. Erosion of blood vessels using Segmentation map
  14. ROP CNNs using Transfer Learning
  15. Results
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Retina showing zones/hours for location and extent of ROP jefferies2016retinopathy
  • Figure 2: Erosion of blood vessels using LWNET vessel map
  • Figure 3: McROP Retcam samples using image domain based Pre-processing
  • Figure 4: McROP Retcam samples using restoration based based Pre-processing
  • Figure 5: McROP Retcam samples using Erosion based Pre-processing