Development and Evaluation of Artificial Intelligence-Based Two-Step Model for Automated Serum Quality Assessment in Clinical Laboratories.

Background: Enhancing pre-analytical QC in clinical laboratories and providing risk alerts is crucial for ensuring accurate test results for patients whose samples are affected by hemolysis, icterus, or lipemia. We evaluated a novel, self-developed artificial intelligence (AI) approach for assessing serum quality using a two-step deep learning YOLOv5-ResNet-50 architecture model and explored its performance under real-world laboratory conditions. Methods: The performance of the two-step model prototype integrated into a Roche Cobas c 701 module (Roche Diagnostics, Shanghai, China) for preprocessing was assessed. By combining serum index testing with manual evaluation, we established a laboratory-specific model for automated sample quality identification. A primary dataset of 21,000 serum samples was partitioned into a training set (80%) and a validation set (20%) for model development and tuning. Subsequently, a separate, independent test dataset of 21,000 samples was used to conduct the final, unbiased evaluation of the model's performance, with statistical analysis. Results: The AI serum model achieved recognition accuracies exceeding 97% for hemolysis and icterus and over 92% for lipemia. In human-machine comparisons, the AI model demonstrated higher consistency (72% exact agreement) than manual assessments (42%-68% exact agreement). Conclusions: The AI-powered serum quality software demonstrated high accuracy in preanalytical serum quality assessment, offering a performance that meets or exceeds that of both automated and manual measurements. This tool can enhance laboratory efficiency and reduce the error risks associated with subjective visual evaluations.