Automated image analysis – the future of X-rays
TB is still considered a leading cause of death and a major threat to child health worldwide. Although both TB infection and TB disease can be treated with antibiotics, most children who die of TB are never diagnosed nor treated. In clinical practice, experienced clinicians assess TB by examining chest radiographs (CXR). However, paediatric CXRs present specific challenges compared to those of adults, making it difficult to diagnose TB in children. As an alternative to this, computer-aided diagnostic systems based on artificial intelligence have shown comparable performance to TB readings performed by experienced radiologists, which could facilitate mass screening for TB and reduce the clinical burden.
In this context, the Biomedical Image Technologies (BIT) research group from the Polytechnic University of Madrid (UPM) is participating in the development of a system to aid in the diagnosis of paediatric TB through the analysis of CXRs. The objective of this collaboration is to obtain radiographs from the hospitals participating in the consortium in order to train and evaluate the algorithms that compose the diagnostic aid system.
Through other collaborations, the Manhiça Health Research Center, the Barcelona Institute of Global Health and the Spanish Paediatric Tuberculosis Network, BIT-UPM has created a carefully labelled cohort with which we have been able to start working on the development of the artificial intelligence systems.
In addition, a telemedicine platform has been created for the evaluation of radiographs collected by specialized radiologists that is currently being used to label more than 600 radiographs. This platform will also be used for the evaluation of radiographs collected in the future through the participation of the BIT-UPM group in the RISE INNOVA4TB consortium.
Recent work has yielded preliminary results in the paediatric cohort mentioned above, with sensitivity and specificity levels approaching those indicated in the WHO Target Product Profile (TPP) minimum criteria (intended for adults) for a triage test for active pulmonary TB. Although these minimum criteria are not intended for children, approaching these levels indicates that we are moving in the right direction.