Menu Close

AI procedure in Chest X-Ray screening (2021)


Chest X-ray (CXR) is a screening tool for tuberculosis (TB) and other lung diseases, used for annual health check-ups and before surgeries. Although TB occurs in every part of the world, Thailand is classified by the World Health Organization (WHO) as one of the 22 countries with the highest TB burden. In Thailand, despite performing more than 1M CXR every year, only about 20% of the CXRs are interpreted by radiologists with 9% of human error rate which leads to misdiagnosis. The development of an artificial intelligent (AI) tool for CXR screening will act as a triage measure to screen CXR before further classification by radiologists as well as reducing the time used for interpretation.

This project will build a Proof of Concept (PoC) for an AI software that would serve as a successor to an AI facility at Songklanagarind Hospital aiming to reduce false negatives diagnosis. This project will use 10,000 CXR records from patients with and without lesions collected in 2018-2019 at the Songklanagarind Hospital. 

The main objectives of this project can be summarised as follow:

  1. Understand CXR screening processing needs in the Thai population.
  2. Address one of the GCRF challenge area(s) as well as WHO global challenges by screen TB in lower and middle-income countries. 
  3. Build a PoC  for an AI tool for CXR screening for a number of end users including researchers and clinicians. The researchers at the PSU will work on a feature extraction approach to be compared with a DL model developed by UoE. Both models could be integrated to enhance the efficiency of detecting lesions on CXRs.
  4. Data sharing  between the partner institutions. Songklanagarind Hospital in collaboration with PSU, the associated partner, will provide CXRs and radiologists’ annotations.
  5. Knowledge transfer between UoE and PSU to study the viability of the development of a CXR screening facility for the Songklanagarind Hospital

Project members

Team members


The project is funded through a GCRF@Essex Research Grant No G026.