Doctors Identify Hemorrhage Better during Chart Review when Assisted by Artificial Intelligence.

OBJECTIVES: This study evaluated if medical doctors could identify more hemorrhage events during chart review in a clinical setting when assisted by an artificial intelligence (AI) model and medical doctors' perception of using the AI model. METHODS: To develop the AI model, sentences from 900 electronic health records were labeled as positive or negative for hemorrhage and categorized into one of 12 anatomical locations. The AI model was evaluated on a test cohort consisting of 566 admissions. Using eye-tracking technology, we investigated medical doctors' reading workflow during manual chart review. Moreover, we performed a clinical use study where medical doctors read two admissions with and without AI assistance to evaluate performance when using and perception of using the AI model. RESULTS: The AI model had a sensitivity of 93.7% and a specificity of 98.1% on the test cohort. In the use studies, we found that medical doctors missed more than 33% of relevant sentences when doing chart review without AI assistance. Hemorrhage events described in paragraphs were more often overlooked compared with bullet-pointed hemorrhage mentions. With AI-assisted chart review, medical doctors identified 48 and 49 percentage points more hemorrhage events than without assistance in two admissions, and they were generally positive toward using the AI model as a supporting tool. CONCLUSION: Medical doctors identified more hemorrhage events with AI-assisted chart review and they were generally positive toward using the AI model.

Automatic Detection of Abnormalities and Grading of Diabetic Retinopathy in 6-Field Retinal Images: Integration of Segmentation Into Classification.

Purpose: Classification of diabetic retinopathy (DR) is traditionally based on severity grading, given by the most advanced lesion, but potentially leaving out relevant information for risk stratification. In this study, we aimed to develop a deep learning model able to individually segment seven different DR-lesions, in order to test if this would improve a subsequently developed classification model. Methods: First, manual segmentation of 34,075 different DR-lesions was used to construct a segmentation model, with performance subsequently compared to another retinal specialist. Second, we constructed a 5-step classification model using a data set of 31,325 expert-annotated retinal 6-field images and evaluated if performance was improved with the integration of presegmentation given by the segmentation model. Results: The segmentation model had higher average sensitivity across all abnormalities compared to the retinal expert (0.68 and 0.62) at a comparable average F1-score (0.60 and 0.62). Model sensitivity for microaneurysms, retinal hemorrhages and intraretinal microvascular abnormalities was higher by 42.5%, 8.8%, and 67.5% and F1-scores by 15.8%, 6.5%, and 12.5%, respectively. When presegmentation was included, grading performance increased by 29.7%, 6.0%, and 4.5% for average per class accuracy, quadratic weighted kappa, and multiclass macro area under the curve, with values of 70.4%, 0.90, and 0.92, respectively. Conclusions: The segmentation model matched an expert in detecting retinal abnormalities, and presegmentation substantially improved accuracy of the automated classification model. Translational Relevance: Presegmentation may yield more accurate automated DR grading models and increase interpretability and trust in model decisions.

Deep Learning-Based Algorithms in Screening of Diabetic Retinopathy: A Systematic Review of Diagnostic Performance.

TOPIC: Diagnostic performance of deep learning-based algorithms in screening patients with diabetes for diabetic retinopathy (DR). The algorithms were compared with the current gold standard of classification by human specialists. CLINICAL RELEVANCE: Because DR is a common cause of visual impairment, screening is indicated to avoid irreversible vision loss. Automated DR classification using deep learning may be a suitable new screening tool that could improve diagnostic performance and reduce manpower. METHODS: For this systematic review, we aimed to identify studies that incorporated the use of deep learning in classifying full-scale DR in retinal fundus images of patients with diabetes. The studies had to provide a DR grading scale, a human grader as a reference standard, and a deep learning performance score. A systematic search on April 5, 2018, through MEDLINE and Embase yielded 304 publications. To identify potentially missed publications, the reference lists of the final included studies were manually screened, yielding no additional publications. The Quality Assessment of Diagnostic Accuracy Studies 2 tool was used for risk of bias and applicability assessment. RESULTS: By using objective selection, we included 11 diagnostic accuracy studies that validated the performance of their deep learning method using a new group of patients or retrospective datasets. Eight studies reported sensitivity and specificity of 80.28% to 100.0% and 84.0% to 99.0%, respectively. Two studies report accuracies of 78.7% and 81.0%. One study provides an area under the receiver operating curve of 0.955. In addition to diagnostic performance, one study also reported on patient satisfaction, showing that 78% of patients preferred an automated deep learning model over manual human grading. CONCLUSIONS: Advantages of implementing deep learning-based algorithms in DR screening include reduction in manpower, cost of screening, and issues relating to intragrader and intergrader variability. However, limitations that may hinder such an implementation particularly revolve around ethical concerns regarding lack of trust in the diagnostic accuracy of computers. Considering both strengths and limitations, as well as the high performance of deep learning-based algorithms, automated DR classification using deep learning could be feasible in a real-world screening scenario.