Leveraging Transfer Learning from Indian Patient Metadata for Deep Learning-Based Chest X-ray Interpretation

Deep learning methods have become very popular as AI has improved. They are used to make strong classification models that work well in many areas, such as medical diagnosis jobs. It is suggested in this study that the CNN model (Convolutional Neural Network) be used to sort Chest Radiological Society of North America (RSNA) Pneumonia databases with X-ray pictures. The study also tries to try out different ways to get the same RSNA test results with the limited computing power techniques to the methods that have been used in the last few years. What the suggested method is based on a CNN that isn’t too complicated and the use of transfer learning algorithms like Xception and Inception V3/V4. NetB7 is efficient. The study also tries to get the same RSNA standard scores using the limited computer resources by trying out different ways to use the methods that have already been put in place in the last few years. RSNA’s standard MAP score is 0.25, but when the Mask RCNN model is used on A randomly chosen group of 3017 Indian people and picture enhancement led to a MAP score of 0.15. At the same time, the YoloV3 the MAP score was 0.32 when no hyperparameters were tuned, but the loss keeps going down. Running if you run the model more times, you might get better results.