Application of deep learning network in clinical diagnosis ofactive pulmonary tuberculosis based on CD161

Abstract

Objective To explore the diagnostic value of cell surface molecule CD161 by flow cytometrytechnology, and to establish deep learning networks that can distinguish sputum smear-negative pulmonary tuber‐culosis, sputum smear-negative IGRA positive/negative pulmonary tuberculosis and pneumonia patients. Methods The proportions of lymphocytes, monocytes and CD161-positive lymphocytes were detected by flow cytometry,and used to construct classification model using deep learning networks. Results The tests on the deep learningnetworks showed that the ratios of three cell populations were able to distinguish sputum smear-negative tubercu‐losis, sputum smear-negative IGRA-positive tuberculosis, and sputum smear-negative IGRA-positive tuberculosisfrom pneumonia patients. Conclusion Based on CD161-flow cytometry technique might be used as an auxiliarydiagnostic method to make a preliminary distinction between sputum smear-negative, sputum smear-negative IG‐RA-positive/negative tuberculosis and pneumonia patients, to improve detection rate of sputum smear-negative tu‐berculosis patients and guide clinical treatment in advance.

Keywords

• active pulmonary tuberculosis
• pneumonia; cd161
• sputum smear
• deep learning network