Comparison of Artificial Intelligence Automated Quantification and Visual Assessment in Optical Coherence Tomography Images (CONTEST)

Background: Optical coherence tomography (OCT) with artificial intelligence (AI) has been developed. Aims: The study aimed to evaluate the differences between AI-quantified and visual assessments. Methods: Patients scheduled for OCT-guided percutaneous coronary intervention between September 2021 and October 2022 were included. AI-quantified OCT pullback and another visually assessed pullback were acquired. The two pullbacks were manually analyzed offline after the procedure (visual analysis 1 and 2). The calcium parameters and vessel size quantified by AI in the corresponding cross-sections assessed in visual analysis 1 were selected (AI-quantified analysis). Intraclass correlation coefficients (ICC) were computed, and Bland−Altman analysis was performed. Subsequent procedural strategies based on the AI-quantified and visual assessments were compared. Results: The concordance of calcified plaque presence between AI-quantified analysis and visual analysis 1 or 2 was 83%. The ICC of calcium length between AI-quantified analysis and visual analysis 1 or 2 was above 0.80. However, the ICC values were below 0.75 in maximum calcium angle, the mean calcium angle, and the maximum calcium thickness. The Bland−Altman plots indicated that the calcium angle and thickness quantified by AI were greater than those assessed visually. The ICC for EEL and lumen diameters at proximal and distal references were all approximately above 0.90. There was no difference between the strategies based on the AI-quantified and visual assessments. Conclusions: AI-quantified assessments tended to overestimate calcium presence and parameters compared to visual assessments. However, vessel sizes demonstrated excellent correspondence. Strategies were similarly developed based on both assessments.