PROJECT TITLE :

Automatic Detection of Tuberculosis in Chest Radiographs Using a Combination of Textural, Focal, and Shape Abnormality Analysis

ABSTRACT:

Tuberculosis (TB) may be a common disease with high mortality and morbidity rates worldwide. Automatic systems to detect TB on chest radiographs (CXRs) can improve the potency of diagnostic algorithms for pulmonary TB. The numerous manifestation of TB on CXRs from totally different populations needs a system that can be custom-made to accommodate different varieties of abnormalities. A laptop aided detection (CAD) system was developed that combines many subscores of supervised subsystems detecting textural, form, and focal abnormalities into one TB score. A general framework was developed to mix an arbitrary variety of subscores: subscores were normalized, collected in a feature vector and then combined using a supervised classifier into one combined score. The method was evaluated on 2 databases, both consisting of two hundred digital CXRs, from: (A) Western high-risk group screening, (B) TB suspect screening in Africa. The subscores and combined score were compared to (1) an external, non-radiological, reference and (two) a radiological reference determined by a person's expert. Performance was measured using Receiver Operator Characteristic (ROC) analysis. Completely different subscores performed best in the two databases. The combined TB score performed better than the individual subscores, aside from the external reference in database B. The performances of the freelance observer were slightly on top of the combined TB score. Compared to the external reference, variations in performance between the combined TB score and therefore the freelance observer were not significant in each databases. Supervised combination to compute an overall TB score allows for a necessary adaptation of the CAD system to completely different settings or totally different operational requirements.