Quality and content analysis of fundus images using deep learning

• Published in: Computers in biology and medicine Vol. 108; pp. 317 - 331
• Main Authors: Chalakkal, Renoh Johnson, Abdulla, Waleed Habib, Thulaseedharan, Sinumol Sukumaran
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.05.2019; Elsevier Limited
• Subjects: Algorithms; Classification; Content analysis; Deep learning; Diabetes; Diabetic retinopathy; Direct reduction; Fundus images; Homogeneity; Image analysis; Image classification; Image processing; Image quality; Medical screening; Methods; Neural networks; Performance evaluation; Performance measurement; Qualitative research; Quality; Quality assessment; Retina; Retinal image quality analysis; Retinal images; Sharpness; Transfer learning; New Zealand; Deep learning; Fundus images; Transfer learning; Retinal image quality analysis
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2019.03.019

Automatic retinal image analysis has remained an important topic of research in the last ten years. Various algorithms and methods have been developed for analysing retinal images. The majority of these methods use public retinal image databases for performance evaluation without first examining the retinal image quality. Therefore, the performance metrics reported by these methods are inconsistent. In this article, we propose a deep learning-based approach to assess the quality of input retinal images. The method begins with a deep learning-based classification that identifies the image quality in terms of sharpness, illumination and homogeneity, followed by an unsupervised second stage that evaluates the field definition and content in the image. Using the inter-database cross-validation technique, our proposed method achieved overall sensitivity, specificity, positive predictive value, negative predictive value and accuracy of above 90% when tested on 7007 images collected from seven different public databases, including our own developed database—the UoA-DR database. Therefore, our proposed method is generalised and robust, making it more suitable than alternative methods for adoption in clinical practice. •Pre-trained deep convolutional neural networks (DCNN) using transfer learning detects low quality and outlier images.•Unsupervised level two classification helps in robust detection of medically suitable retinal image (MSRI).•Transfer learning using fine-tuned DCNN pre-trained on millions of images, negotiates large labelled dataset requirement.•Overall sensitivity, specificity, positive predictive value, negative predictive value and accuracy achieved is above 90%.•7007 images from seven different public databases are used for validation.