Automatic detection of mycobacterium tuberculosis using artificial intelligence

Xiong, Yan; Ba, Xiaojun; Hou, Ao; Zhang, Kaiwen; Longsen, Chen; Li, Ting

doi:10.21037/jtd.2018.01.91

Cited by 95 publications

(66 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The investigators used a Gaussian model to apply various techniques that filter background interference, with a precision of >90%. Additionally, a study in China used a convolutional neural network model to recognize tuberculosis bacillus; this study used 201 samples for the test set, and the results were confirmed by pathologists, finding a sensitivity of 97.94% and a specificity of 83.65% (17).…”

Section: Discussionmentioning

confidence: 84%

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Sua

Bolaños

Maya

et al. 2020

Preprint

View full text Add to dashboard Cite

Background Tuberculosis is the first infection-related cause of death worldwide. Early diagnosis of paucibacillary tuberculosis represents a challenge, even with direct tissue examination. Digital pathology allows the digital analysis of tissues to identify microorganisms. We aim to develop a program to detect and quantify typical and atypical mycobacteria in paraffin-embedded Ziehl–Neelsen-stained tissues. Methods Program development: The building of the program, named Pat-Scan, included pathology, systems engineering, and scientific applications. The iScan Coreo Au scanner® was used, and 9 variables were adjusted: Temp Directory Path, Output Directory Path, Server Path, Focus Approach, Focus Mode, AOI Detection Approach, Scan at, No. of Z Layers, and Z Delta. Software parameter settings : Brightness, contrast, sharpness, and red/green/blue. Control module scan and analysis : Ten Ziehl–Neelsen-stained samples were fragmented into 2,000 images and analyzed by a multidisciplinary team to validate the reproducibility of the bacilli images in the tissue, as detected by the software. Results Pat-Scan included software and a scanner that were used to detect and quantify bacilli in paraffin-embedded Ziehl–Neelsen-stained tissues. HD quality image segmentation was performed, and nine planes of the Z-axis were scanned with a 1-micron distance between planes. Image magnification: 40x–80x. Scan time: 10–12 minutes. All samples containing mycobacteria were successfully analyzed by the scanner, and the bacilli could be detected; these results were validated by expert pathologists by microscopy examination, and the presence of bacilli was confirmed in all cases. Conclusions Pat-Scan allowed the identification and quantification of mycobacteria in paraffin-embedded Ziehl–Neelsen-stained tissues, offering a reproducible diagnostic method that reduces the time for diagnosis and does not affect precision. Further validation is needed for application in the clinical setting.

show abstract

Section: Discussionmentioning

confidence: 84%

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Sua

Bolaños

Maya

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

Section: Discussionmentioning

confidence: 84%

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Sua

Bolaños

Maya

et al. 2020

Preprint

View full text Add to dashboard Cite

BackgroundTuberculosis is the first infection-related cause of death worldwide. Early diagnosis of paucibacillary tuberculosis represents a challenge, even with direct tissue examination. Digital pathology allows the digital analysis of tissues to identify microorganisms. We aim to develop a program to detect and quantify typical and atypical mycobacteria in paraffin-embedded Ziehl–Neelsen-stained tissues.MethodsProgram development: The building of the program, named Pat-Scan, included pathology, systems engineering, and scientific applications. The iScan Coreo Au scanner® was used, and 9 variables were adjusted: Temp Directory Path, Output Directory Path, Server Path, Focus Approach, Focus Mode, AOI Detection Approach, Scan at, No. of Z Layers, and Z Delta. Software parameter settings: Brightness, contrast, sharpness, and red/green/blue. Control module scan and analysis: Ten Ziehl–Neelsen-stained samples were fragmented into 2,000 images and analyzed by a multidisciplinary team to validate the reproducibility of the bacilli images in the tissue, as detected by the software.ResultsPat-Scan included software and a scanner that were used to detect and quantify bacilli in paraffin-embedded Ziehl–Neelsen-stained tissues. HD quality image segmentation was performed, and nine planes of the Z-axis were scanned with a 1-micron distance between planes. Image magnification: 40x–80x. Scan time: 10–12 minutes. All samples containing mycobacteria were successfully analyzed by the scanner, and the bacilli could be detected; these results were validated by expert pathologists by microscopy examination, and the presence of bacilli was confirmed in all cases.ConclusionsPat-Scan allowed the identification and quantification of mycobacteria in paraffin-embedded Ziehl–Neelsen-stained tissues, offering a reproducible diagnostic method that reduces the time for diagnosis and does not affect precision. Further validation is needed for application in the clinical setting.

show abstract

“…The NIAID extramural portfolio of grants and contracts supports all aspects of TB research, ranging from studying the basic biology of Mycobacterium tuberculosis and its interaction with the host to investigating the various manifestations of TB (pulmonary, extrapulmonary, and latent) in adult and pediatric populations, including HIV-coinfected individuals, to conducting research aimed at developing new health care interventions. [9] Total of 66 MDR TB patients started treatment, from the study population and among them 20 (30%) were resistant to one or more second line drugs including a case of "XDR TB". For treatment only half of the patients The genome sequences including positions outside 23 genes and deep networks for nonlinear classification and dimension reduction and also optimising the number of SPCA/SNMF components can be considered as the future work.…”

Section: Related Workmentioning

confidence: 99%

Prediction of Multi Drug Resistant Tuberculosis using Machine Learning Techniques

Lokeshkumar,

kumar

et al. 2019

IJEAT

View full text Add to dashboard Cite

 Abstract: Mycobacterium Tuberculosis bacteria is the primary cause for Tuberculosis. TB is one of the main reasons of mortality around the world. Multi Drug Resistant Tuberculosis (MDR-TB) is a type of tuberculosis bacteria which are resistant to anti-TB drugs, drugs like isoniazid (INH) and rifampin (RMP). Different Machine learning approaches has been widely applied to predict MDR TB. Here, we review different Machine Learning Approaches to predict MDR-TB. Different feature estimation methods, execution of distinct machine learning models also have been explored. Additionally, the utilization of the distinctive machine learning system models for distinguishing the dis-functionalities of MDR-TB in the recent decades has been talked about.

show abstract

Automatic detection of mycobacterium tuberculosis using artificial intelligence

Cited by 95 publications

References 1 publication

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Detection of Mycobacteria in Paraffin-Embedded Ziehl–Neelsen-Stained Tissues using Digital Pathology

Prediction of Multi Drug Resistant Tuberculosis using Machine Learning Techniques

Contact Info

Product

Resources

About