Multilayer Scanning Enhances Sensitivity Of Artificial Intelligence-aided Mycobacterium Tuberculosis Detection

Xiong Yan, Hou Ao, Li Ting, Chen Longsen, Chen Lifang, Lai Lili

Abstract: In the study of automatic detection of Mycobacterium Tuberculosis (TB) using artificial intelligence, 201 samples (108 positive cases and 93 negative cases) were collected as a test set and used to examine TB-AI and TB-AI achieved 97.94% sensitivity and 83.65% specificity. However, with single-layer scanning, some Mycobacterium TBs are blurred due to the defocus. As a result, slides with blurred TB pixels may not be detected as positive In this paper a new test of TB-AI with three-layer scanning was conducted on 189 positive cases reported by medical doctors with microscope. Comparing to the ordinary single-layer scanned slides, additional 6 out of 189 cases (3.2%) were detected.

Keywords: Mycobacterium Tuberculosis (TB), Acid-fast Stainin, Multilayer Scanning, Artificial Intelligence (AI), Auxiliary Diagnosis

Author Affiliation: Peking University First Hospital, semptian