Artificial Intelligence-Assisted Diagnosis of Endobronchial Tuberculosis

With the development of artificial intelligence (AI) in the field of diagnosis, AI-assisted diagnosis has attracted much attention. However, hospitals have no AI diagnosis system for endobronchial tuberculosis (EBTB). To build an AI diagnosis system for EBTB, a high-quality EBTB medical image database and an efficient ResNet34-MH network model that can identify and classify EBTB were built. The traditional 3 × 3 convolution of Layer4 in the original ResNet34 model is replaced with a Multi-Head Self-Attention mechanism (MHSA) module, allowing the model to extract local and global image features. The accuracy of the AI diagnosis system for EBTB was 87.5%, and the time of diagnosis was 41ms.