Development of an Optimized Pneumonia Classification Model from Chest X-ray Images

Pneumonia is a lung infection that leads to the accumulation of fluids in the lungs. Pneumonia impacts 7% of the world population and early detection is necessary for effective treatment. The diagnosis is done by radiologists examining the chest X-rays and determining the abnormalities in the image. The low-quality image and shortage of time in the examination can often lead to making bad predictions, therefore Computer Aided Diagnosis is proposed to make better decisions as it enhances the image and helps in prediction. The proposed model utilizes openly available Kaggle dataset for the development of an effective classifier. Deep Learning segmentation models are used to generate the binary masks. The generated mask was utilized to extract features using the python library pyradiomics. 93 features were generated, and 5 classification models were trained with hyperparameter tuning. The best classifier was used as the base model for making feature selecting using conventional feature selection models. Feature selection helps in selecting the best subset of features which helps in making better predictions. F1-score, precision, recall, and accuracy were taken as the evaluation parameters and based on the analysis Random Forest along with correlation-based feature selection produced the best result. The result obtained was used for the underlying working model for web app. The web app receives the input as an image and produces output as whether a person as pneumonia or not. In conclusion better CAD system was built using all the above-mentioned procedures.