An Efficient Approach of Pneumonia Detection using Transfer Learning Models, RCNN and FAST RCNN

Computer vision can be used in health care for identifying diseases. The bacterial, fungal, or viral inflammation and fluid in one or both lungs are called pneumonia. Therefore, building a model for pneumonia detection is required through chest X-ray such that lung opacities on chest radiographs should be automatically detected by the algorithm. Pneumonia is diagnosed with the help of chest X-rays by qualified radiotherapists. Rarely, pneumonia can be mistaken with other diseases due to similarity of the symptoms. Therefore, designing a deep learning algorithm to detect pneumonia is accomplished by predicting the bounding box co-ordinates on lung opacities to direct the clinicians. The basic CNN model is designed for achieving this and can be tuned with different hyper parameters to achieve the best AUC score. Implementing the model and improving the performance through hyper-parameter tuning was done with transfer learning techniques such as RESNET, DENSENET, InceptionV3, U-Net and Mask-RCNN, Faster RCNN. Thus, an automated pneumonia detection mechanism will help in augmenting the efficacy of clinical decision making while reducing human interventions at the same time.