Machine-assisted interpretation of auramine stains substantially increases through-put and sensitivity of microscopic tuberculosis diagnosis
| Autor: | L. Horvath, S. Hänselmann, H. Mannsperger, S. Degenhardt, K. Last, S. Zimmermann, I. Burckhardt |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
0301 basic medicine
Microbiology (medical) medicine.medical_specialty Tuberculosis 030106 microbiology Immunology Microbiology 03 medical and health sciences Deep Learning Tuberculosis diagnosis BACTERIAL INFECTIOUS DISEASES Microscopy medicine Image Processing Computer-Assisted Humans Coloring Agents Retrospective Studies Staining and Labeling business.industry Sputum Mycobacterium tuberculosis Automated microscopy medicine.disease 030104 developmental biology Infectious Diseases Microscopy Fluorescence Benzophenoneidum Radiology medicine.symptom business |
| Zdroj: | Tuberculosis (Edinburgh, Scotland). 125 |
| ISSN: | 1873-281X |
| Popis: | Of all bacterial infectious diseases, infection by Mycobacterium tuberculosis poses one of the highest morbidity and mortality burdens on humans throughout the world. Due to its speed and cost-efficiency, manual microscopy of auramine-stained sputum smears remains a crucial first-line detection method. However, it puts considerable workload on laboratory staff and suffers from a limited sensitivity. Here we validate a scanning and analysis system that combines fully-automated microscopy with deep-learning based image analysis. After automated scanning, the system summarizes diagnosis-relevant image information and presents it to the microbiologist in order to assist diagnosis. We tested the benefit of the automated scanning and analysis system using 531 slides from routine workflow, of which 56 were from culture positive specimen. Assistance by the scanning and analysis system allowed for a higher sensitivity (40/56 positive slides detected) than manual microscopy (34/56 positive slides detected), while greatly reducing manual slide-analysis time from a recommended 5–15 min to around 10 s per slide on average. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect