Segmentation of X-ray coronary angiography with an artificial intelligence deep learning model: Impact in operator visual assessment of coronary stenosis severity.

BACKGROUND: Visual assessment of the percentage diameter stenosis (%DSVE ) of lesions is essential in coronary angiography (CAG) interpretation. We have previously developed an artificial intelligence (AI) model capable of accurate CAG segmentation. We aim to compare operators' %DSVE in angiography versus AI-segmented images. METHODS: Quantitative coronary analysis (QCA) %DS (%DSQCA ) was previously performed in our published validation dataset. Operators were asked to estimate %DSVE of lesions in angiography versus AI-segmented images in separate sessions and differences were assessed using angiography %DSQCA as reference. RESULTS: A total of 123 lesions were included. %DSVE was significantly higher in both the angiography (77% ± 20% vs. 56% ± 13%, p < 0.001) and segmentation groups (59% ± 20% vs. 56% ± 13%, p < 0.001), with a much smaller absolute %DS difference in the latter. For lesions with %DSQCA of 50%-70% (60% ± 5%), an even higher discrepancy was found (angiography: 83% ± 13% vs. 60% ± 5%, p < 0.001; segmentation: 63% ± 15% vs. 60% ± 5%, p < 0.001). Similar, less pronounced, findings were observed for %DSQCA < 50% lesions, but not %DSQCA > 70% lesions. Agreement between %DSQCA /%DSVE across %DSQCA strata (<50%, 50%-70%, >70%) was approximately twice in the segmentation group (60.4% vs. 30.1%; p < 0.001). %DSVE inter-operator differences were smaller with segmentation. CONCLUSION: %DSVE was much less discrepant with segmentation versus angiography. Overestimation of %DSQCA < 70% lesions with angiography was especially common. Segmentation may reduce %DSVE overestimation and thus unwarranted revascularization.

Coronary X-ray angiography segmentation using Artificial Intelligence: a multicentric validation study of a deep learning model.

INTRODUCTION: We previously developed an artificial intelligence (AI) model for automatic coronary angiography (CAG) segmentation, using deep learning. To validate this approach, the model was applied to a new dataset and results are reported. METHODS: Retrospective selection of patients undergoing CAG and percutaneous coronary intervention or invasive physiology assessment over a one month period from four centers. A single frame was selected from images containing a lesion with a 50-99% stenosis (visual estimation). Automatic Quantitative Coronary Analysis (QCA) was performed with a validated software. Images were then segmented by the AI model. Lesion diameters, area overlap [based on true positive (TP) and true negative (TN) pixels] and a global segmentation score (GSS - 0 -100 points) - previously developed and published - were measured. RESULTS: 123 regions of interest from 117 images across 90 patients were included. There were no significant differences between lesion diameter, percentage diameter stenosis and distal border diameter between the original/segmented images. There was a statistically significant albeit minor difference [0,19 mm (0,09-0,28)] regarding proximal border diameter. Overlap accuracy ((TP + TN)/(TP + TN + FP + FN)), sensitivity (TP / (TP + FN)) and Dice Score (2TP / (2TP + FN + FP)) between original/segmented images was 99,9%, 95,1% and 94,8%, respectively. The GSS was 92 (87-96), similar to the previously obtained value in the training dataset. CONCLUSION: the AI model was capable of accurate CAG segmentation across multiple performance metrics, when applied to a multicentric validation dataset. This paves the way for future research on its clinical uses.