Abstract
BACKGROUND: Tuberculosis (TB) is a prevalent infectious disease that infects about a quarter of the world’s population and has become one of the leading causes of death in the world’s population. Early diagnosis and treatment are crucial for preventing the spread of the disease and improving the patient’s prognosis. Mycobacterium tuberculosis microscopy can provide a basis for TB disease diagnosis, the accuracy of which is influenced by the doctor’s experience. With the development of artificial intelligence technology, deep learning can provide automated bacteriological diagnostic assistance for doctors. This study aims to develop an efficient and accurate automated detection method for Mycobacterium tuberculosis based on a cloud computing platform utilizing deep learning algorithms. METHODS: The study incorporated 9963 annotated data from 1265 acid-fast stained smears on the Kaggle database. It used the deep learning algorithm YOLOv5 on Jiutian platform to extract and analyze features from the acid-fast stained positive regions and build a disease detection model. RESULTS: The results indicate that the model demonstrates rapid detection of Mycobacterium tuberculosis, with a maximum mAP(50) of 94.80%, an accuracy of 93.72%, and a recall of 91.24%. CONCLUSION: The study is expected to provide accurate and efficient aid for doctors in diagnosing TB.