Algorithm of cancer diagnosis based on artificial intelligence

Cancer, the second leading cause of death globally, necessitates early screening for improved patient survival. This study introduces an innovative approach utilizing T-cell receptors (TCRs) derived from peripheral blood RNA sequencing as diagnostic biomarkers. We developed a comprehensive system for disease diagnosis and feature extraction by integrating three distinct multidimensional datasets and employing diverse machine learning algorithms. Algorithmic optimization involved refining prediction thresholds enhances accuracy. We establish a robust multi-classification framework. Successfully applied to cancer identification tasks, the system demonstrated exceptional performance: an area under the curve (AUC) of 99.8% and an accuracy of 98.4%. Further analysis employing multi-threshold modeling on two-dimensional liver cancer data revealed that models constructed within the cutoff2 to cutoff4 range consistently achieved perfect scores (AUC=1.000, Accuracy=1.000). This indicates outstanding classification performance inherent in data within this interval. Subsequent parameter optimization also yielded additional performance gains in the models.