Automatic interpretation and clinical evaluation for fundus fluorescein angiography images of diabetic retinopathy patients by deep learning

Abstract

Background/aimsFundus fluorescein angiography (FFA) is an important technique to evaluate diabetic retinopathy (DR) and other retinal diseases. The interpretation of FFA images is complex and time-consuming, and the ability of diagnosis is uneven among different ophthalmologists. The aim of the study is to develop a clinically usable multilevel classification deep learning model for FFA images, including prediagnosis assessment and lesion classification.MethodsA total of 15 599 FFA images of 1558 eyes from 845 patients diagnosed with DR were collected and annotated. Three convolutional neural network (CNN) models were trained to generate the label of image quality, location, laterality of eye, phase and five lesions. Performance of the models was evaluated by accuracy, F-1 score, the area under the curve and human-machine comparison. The images with false positive and false negative results were analysed in detail.ResultsCompared with LeNet-5 and VGG16, ResNet18 got the best result, achieving an accuracy of 80.79%–93.34% for prediagnosis assessment and an accuracy of 63.67%–88.88% for lesion detection. The human-machine comparison showed that the CNN had similar accuracy with junior ophthalmologists. The false positive and false negative analysis indicated a direction of improvement.ConclusionThis is the first study to do automated standardised labelling on FFA images. Our model is able to be applied in clinical practice, and will make great contributions to the development of intelligent diagnosis of FFA images.

Investigating Novel Thiophene Carbaldehyde Based Thiazole Derivatives as Potential Hits for Diabetic Management: Synthesis, In Vitro and In Silico Approach

“…[36,37] Controlling the catalytic activity of the enzyme αglucosidase (EC.3.2.1.20), located in the small intestine, is one strategy for treating this condition. [38,39] The α-1-4 bond linkage in starch or oligosaccharides is broken down by this enzyme into monosaccharides like glucose. [40,41] Therefore, α-glucosidase suppression can aid in preventing post-prandial hyperglycemia and its related consequences.…”

“…Improved treatments for this chronic illness are urgently needed because it is predicted that 300 million individuals worldwide would have diabetes mellitus by the year 2025 [36,37] . Controlling the catalytic activity of the enzyme α‐glucosidase (EC.3.2.1.20), located in the small intestine, is one strategy for treating this condition [38,39] . The α‐1‐4 bond linkage in starch or oligosaccharides is broken down by this enzyme into monosaccharides like glucose [40,41] .…”

A classification algorithm based on improved meta learning and transfer learning for few‐shot medical images

“…In recent years, the development of deep learning has promoted the research of medical image classification [1], segmentation [2], registration [3], and other aspects, and has achieved some remarkable results. For example, the two-stage deep learning system proposed by Jin et al [4] has effectively improved the average accuracy of the traditional deep learning model in the task of grading the severity of the epiretinal membranes; Gao et al [5] used convolutional network to study the automatic standardized labelling of fundus fluorescein angiography images, which also effectively improved the classification accuracy; Zhuang et al [6] proposed a weakly supervised similarity evaluation network, which has excellent performance in the similarity analysis of lung high-resolution CT images. And it has played a good auxiliary medical role in many clinical fields such as pathology [7].…”

“…[4] has effectively improved the average accuracy of the traditional deep learning model in the task of grading the severity of the epiretinal membranes; Gao et al. [5] used convolutional network to study the automatic standardized labelling of fundus fluorescein angiography images, which also effectively improved the classification accuracy; Zhuang et al. [6] proposed a weakly supervised similarity evaluation network, which has excellent performance in the similarity analysis of lung high‐resolution CT images.…”

Segmentation of nucleus based on dynamic convolution and deep features of stain distribution

“…Currently, traditional methods such as active contour models, watershed algorithms and graph cut algorithms [2][3][4][5], which are usually based on simple features and thresholding strategies, cannot deal with pathological images with drastic variations of nuclei density, blurred boundaries, overlapping targets, mitosis nucleus and sources from different organs. In recent years, deep learning methods have been widely applied to medical image segmentation due to the ability to learn representations of medical images with multiple levels of abstraction [6][7][8][9][10][11]. For small medical image set, Unet which fuses multi-level and multi-scale features through skip connections, achieves excellent performance in segmenting HeLa cells in differential interference contrast microscopy [12].…”