A miniaturized liver function detection system with machine learning enhancing strategy

Abstract

Serum alanine aminotransferase (ALT) is one of the most sensitive indicators of liver function and is crucial in diagnosing acute liver injury (ALI). However, its widespread clinical application is limited due to expensive equipment, detection delays, and high technical requirements, especially in resource-limited areas. This paper presents a miniaturized liver function detection system with machine learning enhancing strategy to address these challenges. The system integrates a quantitative detection algorithm based on grayscale processing and a semi-quantitative detection algorithm based on a convolutional neural network (CNN). The CNN model consists of four block units, each comprising a convolutional layer, an activation function, and a pooling layer. The model was validated using the hold-out method. Additionally, the detection instrument employs rapid and stable temperature control at 37 ± 0.4 °C to maintain serum enzyme activity. The system achieves a quantitative limit of detection of 5.47 U/L, with a measurable range of 6∼395 U/L. The semi-quantitative classification intervals are <1 × ULN (Upper limit of normal), 1–5 × ULN, and >5 × ULN. Accurate ALT concentration results can be obtained within 3 min. Compared to results from a fully automated biochemical analyzer, the system exhibits a linear correlation coefficient of 0.9930 for quantitative detection and an accuracy of 96.97 % for semi-quantitative detection. The impact of common interfering substances is less than 8 %, and the coefficient of variation (CV) is consistently below 10 %, demonstrating good reliability. This system offers a low-cost, fast, and accurate solution for liver function detection, providing a convenient and efficient alternative for clinical applications.