Evaluation of two AI software for the diagnosis of TB infection : A study protocol (Preprint)

BACKGROUND Tuberculosis (TB) profile in Malaysia showed an average annual growth rate of 2.23%, with an estimated 92 cases per 100,000 people reported in 2018. CXR remains the best conventional method for the early detection of pulmonary TB infection. The intervention of AI in TB diagnosis could efficiently aid human interpreters and reduce health professionals' work burden. To date, no evaluation of AI studies has been carried out in Malaysia. OBJECTIVE This study aims to determine the diagnostic accuracy and evaluate the performance of Qure.ai and Putra Analytica AI software. METHODS We will conduct a retrospective case-control study in Respiratory Medicine Institute (IPR), Kuala Lumpur Health Clinic and Bandar Botanik Klang Health Clinic. Patients' medical reports on TB investigation will be retrieved by accessing electronic and hardcopy medical records and collecting demographic data. Prior to conducting the study, patients' PTB status will be obtained by identifying MTB culture (reference standard) results in order to create a case and a control group. A total of 2000 CXR images will be retrieved, of which 1000 images will be the case (abnormality). Normal and abnormal CXR will be categorized into film and digital CXR, which will be screened onto the said AI software (index tests). RESULTS Results obtained from the AI software will be compared with the reference standard, and significant statistical analysis will be computed CONCLUSIONS We hope that the findings of this evaluation study will provide sufficient information for stakeholders and to implement AI technology in the medical imaging field for better management of TB in hospital and clinic settings.