Development of an Artificial Intelligence Verification System Model for Clinical Biochemistry Laboratories Using Machine Learning

ABSTRACT Background With the exponential growth of testing samples in clinical biochemistry laboratories, traditional manual verification methods are increasingly challenged by inefficiency and human error. Machine learning (ML) technology offers a promising solution to enhance the accuracy and speed of laboratory report validation. The aim of this study was to develop an artificial intelligence (AI) verification system using ML algorithms to address these challenges. Methods From January to November 2024, a dataset comprising 6382 clinical reports with 38 biochemical parameters and demographic characteristics was extracted from the Roche Infinity system. Three ML algorithms (XGBoost, random forest, and logistic regression) were trained and validated using 5‐fold cross‐validation. Model performance was evaluated based on the areas under the receiver operating characteristic curves, precision, recall, F 1 ‐score, and average precision. In a double‐blind test, experienced laboratory personnel were compared with the AI verification system, and the pass rate, false negative rate, and turnaround time were recorded. Results The areas under the receiver operating characteristic curves of the XGBoost, random forest, and logistic regression training models for the test set were 0.970, 0.985, and 0.742, respectively. The precisions were 0.83, 0.87, and 0.55; recalls were 0.92, 0.95, and 0.63; and F 1 ‐scores were 0.87, 0.91, and 0.59, respectively. The precision‐recall curves showed that the average precisions were 0.937, 0.972, and 0.625, respectively. The AI system achieved a pass rate of 88.14% and a false negative rate of 0.62% in the double‐blind test. The median turnaround time was reduced by 23.6% (from 89 to 68 min), with the 90th percentile improving by 14.3% (from 154 to 132 min). Conclusions The constructed AI verification system performed excellently in terms of accuracy and efficiency. This study provides a new solution for the intelligent verification of clinical biochemistry laboratories and has certain application value.