Deep-Learning-Aided Detection of Mycobacteria in Pathology Specimens Increases the Sensitivity in Early Diagnosis of Pulmonary Tuberculosis Compared with Bacteriology Tests

Abstract

The histopathological diagnosis of mycobacterial infection may be improved by a comprehensive analysis using artificial intelligence. Two autopsy cases of pulmonary tuberculosis, and forty biopsy cases of undetected acid-fast bacilli (AFB) were used to train AI (convolutional neural network), and construct an AI to support AFB detection. Forty-two patients underwent bronchoscopy, and were evaluated using AI-supported pathology to detect AFB. The AI-supported pathology diagnosis was compared with bacteriology diagnosis from bronchial lavage fluid and the final definitive diagnosis of mycobacteriosis. Among the 16 patients with mycobacteriosis, bacteriology was positive in 9 patients (56%). Two patients (13%) were positive for AFB without AI assistance, whereas AI-supported pathology identified eleven positive patients (69%). When limited to tuberculosis, AI-supported pathology had significantly higher sensitivity compared with bacteriology (86% vs. 29%, p = 0.046). Seven patients diagnosed with mycobacteriosis had no consolidation or cavitary shadows in computed tomography; the sensitivity of bacteriology and AI-supported pathology was 29% and 86%, respectively (p = 0.046). The specificity of AI-supported pathology was 100% in this study. AI-supported pathology may be more sensitive than bacteriological tests for detecting AFB in samples collected via bronchoscopy.

A New Artificial Intelligence-Based Method for Identifying Mycobacterium Tuberculosis in Ziehl–Neelsen Stain on Tissue

“…Zaizen et al, have an interesting approach when constructing the testing group: the positive cases were those with proven mycobacteriosis either when the biopsy was performed or during follow-up; based on this perspective, AI-supported pathological diagnosis identified 11 positive cases versus 2 positive cases in classical pathological diagnosis, without AI support [ 19 ]. It is unusual for a pathologist to miss 9 cases from a total of 42 (12.5% sensitivity).…”

“…Its validation is solely made on patches. Xiong et al [ 15 ] present a completely automated method of diagnosis while Yang et al [ 16 ], Pantanowitz et al [ 18 ], and Zaizen et al [ 19 ] developed AI-assisted diagnostic methods as a tool in the hands (and eyes) of pathologists. In Yang et al’s method, the pathologist evaluates a score heatmap superposed on the WSI.…”

“…There are other studies published on automatic AFB detection on tissue using WSIs [ 15 , 16 , 17 , 18 , 19 ]. Analyzing tissue is more difficult than analyzing smears.…”

“…Zaizen et al (2022) [ 19 ] developed an algorithm to detect AFB using a pre-trained HALO AI CNN. The dataset consisted of 506 AFB annotated on two autopsy cases with TB; the negative ones including two types of artifacts (nuclei of type I epithelial cells as well as fibrin and hyaline membrane) originating from 40 negative biopsies.…”

“…Additionally, to prevent the situations when the human examiner is not able to identify few bacilli in a paucibacillar TB, cases with specific TB morphology but with negative ZN stains were excluded. One author had to reclassify several cases [ 15 ] and another reported a very high number of cases identified as positive after AI-assisted examination (seven newly identified cases out of a total of nine positive ones) [ 19 ].…”