Weakly Supervised Confidence Aware Probabilistic CAM multi-Thorax Anomaly Localization Network

Most anatomical information and anomalies are provided by chest X-ray (CXR) images and are sometimes adequate for the early diagnosis. However, by observing the radiographs it is a challenging task to recognize multiple occurring thorax diseases. As a result, there is a trend toward developing deep learning systems to assist radiologists. Motivated by this, we propose a generalized weakly supervised Confidence-Aware Probabilistic Class Activation Map (CAPCAM) classification model that localizes anomalies for 13 different thoracic disease. The CAPCAM used CX-Ultranet as the backbone with the combination of Confidence Aware Network (CAN) and Anomaly Detection Network (ADN) which helps the model to utilise all components of feature extracted therefore eliminating the necessity to train them individually and time taken. We experimentally shown that our proposed CAPCAM method sets a new state-of-the-art benchmark by achieving accuracy in terms of Intersection of Bounding Box (IoBB) in the range of 85% - 94%, and F1 scores in the range of 88% - 90% for all thirteen diseases on publicly available large-scale CXR NIH dataset. The proposed CAPCAM pooling also achieves better results than other state of the art (SOTA) pooling methods.