Research on an artificial intelligence model based on the Deeplabv3+ algorithm for assisting clinical rapid and accurate classification of Candida

Objective: To efficiently and accurately identify chromogenic images of Candida and thereby achieve rapid diagnosis, this study adopts an artificial intelligence(AI) model based on the Deeplabv3+ algorithm. Methods: A total of 167 strains of clinically common Candida were selected, and 1,020 images were collected after chromogenic culture for the construction and performance verification of the Deeplabv3+ model. Results: The test results showed that the AI achieved an identification accuracy of 91.00% for Candida albicans, 94.00% for Candida tropicalis, and 86.00% for Candida glabrata, with an average identification accuracy of 90.33% across the three species. To further verify the performance of the AI technology, three experienced clinical laboratory technicians were selected for simultaneous visual identification testing, and their average identification accuracy was 89.33%. Statistical analysis indicated no significant difference between the AI's identification accuracy and that of human eyes(χ²=0.14, P>0.05). In terms of recognition speed, the average speed of AI recognition was 1.88±0.04 seconds per image, while that of human eye image recognition was 1.93±0.33 seconds per image. No statistically significant difference was observed between the two groups(U=0.45, P>0.05). However, AI image recognition exhibited greater stability than human eye recognition. Additionally, it enabled batch data processing without being affected by multiple interfering factors. Conclusion: The AI model constructed based on the Deeplabv3+ algorithm in this study exhibits efficient and accurate identification capabilities for chromogenic medium images of Candida, demonstrating promising prospects for promotion.