Identification of bacterial infection types in decomposition stages at various temperatures using pathology images and artificial intelligence algorithms

Abstract

Bacterial infections present a significant threat to human health, and accurate identification of infection type is crucial for both clinical and forensic applications. Although traditional diagnostic methods are reliable, they are often time-consuming, require specialized personnel and equipment, and have limited accessibility. Previous studies have demonstrated that pathology images combined with artificial intelligence (AI) algorithms can effectively classify bacterial infections in fresh tissue samples. In this study, we extend this approach to identify bacterial infection types in decomposed tissue under varying temperature conditions. Our findings indicate that decomposition factors, such as putrefaction and autolysis, do not impair model performance. The model exhibits strong classification efficacy across all tested temperatures (25 °C, 37 °C, and 4 °C), demonstrating robustness and generalizability. The overall area under the curve (AUC) values exceeded 0.920 and 0.820 at the patch and whole slide image (WSI) levels, respectively, in the training and testing sets, while surpassing 0.990 at the patch-level in the external validation set. These results confirm that AI-driven computational pathology can reliably distinguish bacterial infection types, even in decomposition states. Our method offers a novel approach for bacterial diagnosis in forensic pathology and supports infection prevention during autopsies.