Pneumonia Detection and Classification with CNN-GAN Augmentation Using Fine Tuning

Pneumonia is a respiratory condition that leads to the enlargement of the lungs due to the accumulation of fluid in the air sacs and alveoli. If left untreated, it can cause damage to one or both lungs. According to a study conducted by the World Health Organization, individuals with weakened immune systems, such as children, babies, and those aged 65 and above, are at a higher risk of contracting this disease. The study considered three types of pneumonia: normal, viral, and bacterial. To diagnose the condition, chest X-rays were used, as they are widely available and cost-effective. The patients' X-ray images were carefully examined during the investigation. We collected pictures from online archives and used GAN (Generative Adversarial Network) to augment the dataset, which improved the effectiveness of our CNN model's training. Next, we conducted tuning optimization to automate the processes in the CNN model. Once the optimization was completed, we tested the model using the dataset and classified the disease using a combination of CNN and GAN with a joint fine-tuning approach and as a result the model achieved an accuracy of 99%. With the help of the proposed approach, pneumonia identification can be accelerated, and even with a limited dataset, the categorization of the illness can be done optimally. This method could potentially be a life-saving tool and offer valuable second opinions for radiologists and other medical professionals, aiding them in the efficient treatment of patients.