Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning

Abstract

The implementation of clinical-decision support algorithms for medical imaging faces challenges with reliability and interpretability. Here, we establish a diagnostic tool based on a deep-learning framework for the screening of patients with common treatable blinding retinal diseases. Our framework utilizes transfer learning, which trains a neural network with a fraction of the data of conventional approaches. Applying this approach to a dataset of optical coherence tomography images, we demonstrate performance comparable to that of human experts in classifying age-related macular degeneration and diabetic macular edema. We also provide a more transparent and interpretable diagnosis by highlighting the regions recognized by the neural network. We further demonstrate the general applicability of our AI system for diagnosis of pediatric pneumonia using chest X-ray images. This tool may ultimately aid in expediting the diagnosis and referral of these treatable conditions, thereby facilitating earlier treatment, resulting in improved clinical outcomes. VIDEO ABSTRACT.

Multilabel multiclass classification of OCT images augmented with age, gender and visual acuity data

“…Despite no addition curation of the data we were able to achive overall acurracy of 86% and exact match of 66.7% on the hold out test set of 12,581 images. Since most of the other studies ether manually curated the training data and label Kermany et al (2018), or performed binary classification Lee et al (2016) our results are encouraging. Class We added other relevant patient information which is often available to clinicians diagosing pathologies on OCT scan such as age, gender and visual acuity.…”

“…Convolutional Neural Networks (CNNs) and other deep neural networks have enabled unprecedented breakthroughs in developing artificial intelligence systems to perform computerassisted diagnosis based on clinical data and several recent studies demonstrate the ability of these algorithms to leverage large clinical datasets to learb how to classify images as exhibiting a pathology Lee et al (2016); Choi (2017); Esteva (2017); Kermany et al (2018). However, most of these studies classify the presence or absence of a single pathology Lee et al (2016).…”

Current state and future prospects of artificial intelligence in ophthalmology: a review

“…His group included 108 312 images (from 4686 patients) to train the MLC and a separate 1000 images (from 633 patients) to validate its capabilities. 38,[46][47][48] Ultimately, as the MLCs perform at a level similar to professional graders, it appears that an MLC created using a CNN could be a useful system for screening for AMD. The MLC achieved a sensitivity of 0.978, a specificity of 0.974 and a AUROC of 0.999 in its task of retinal disease classification.…”

“…46 The MLC stratified the conditions according to need for the patient to be seen, with conditions like choroidal neovascularization and DME classified as 'urgent referrals', and drusen as part of dry AMD classified as a 'routine referral'. Kermany et al developed a diagnostic tool for screening patients with common treatable blinding retinal diseases, including diabetic macular oedema (DME) and AMD.…”

All eyes are on AI

“…The diagnostic performance of AI has been tested most prominently in ophthalmology, because of the widely available and easy-toobtain retinal images and the well-defined standards for the diagnosis of eye diseases. Most recently, a deep-learning algorithm classified age-related macular degeneration and diabetic retinopathy from optical coherence tomography images of the retina 1 . Both of these eye conditions and others (such as glaucoma and macular oedema), have also been automatically assessed by deep learning trained on fundus images [2][3][4] (a retinal fundus image is a photograph of the internal surface at the back of the eye).…”