Automated Report Generation Using Deep Learning

Abstract

Now a days generating report is a bit overhead for Radiologists and receiving that report from any Hospital or laboratory is more than it. To overcome this problem, we are putting some effort. We developed a Web Application containing automated report generation model which will generate the report automatically without any problem. This model will help radiologists in terms of generating reports without analyzing x-rays one by one and putting decision in generating reports. On the other hand, web application will help the patient in terms of receiving report through online system. Not only this Complete information and records will be accessible by the Hospital. Technologies which we are using to develop Model, we are using encoder decoder architecture for generating reports. In encoder architecture we are using the image layer which decoder architecture using LSTM. In the whole model we are using 2 LSTM layers, embedding layer, dropout layer, add layer, image layer etc. NLP playing important role in developing model, libraries which helps us in terms of NLP are Spacy, NLTK and Genism while for image processing we used OpenCV Python, for analyzing Pandas, NumPy, Matplotlib etc helps us a lot. Our Blue score reached about 0.35 on cross validation set while 0.23 on test set using bean search with the width of 7. If we talk about deployment then we used Flask for deployment of model, we made 2 API’s one for single image and other for 2 images (side and front). Web App is developing on the latest technologies like NEXT, Express, MongoDB, React Bootstrap, MUI and styled components. Our Web App contains three panel all are simply accessible by one login. This was a bit analyzing task for us to develop an optimize Login API where all three roles can get their panel in optimize way because this way can consume bit more CPU usage. We made our web app in very maintainable and optimize manner which everyone can easily accessible. We used the professional folder structure which industry is following also we made very optimize and high-quality code. In our web app radiologist can generate the report through their panel. Patient can get the report once radiologist confirms after generating. Lastly Admin can track the records. This project is still under constructure because this is on high scale and containing lot of functionalities which needs some time to fulfill. In conclusion, we used two backends one (Flask) for serving model and other (Express) for interacting Database (MongoDB). Furthermore, the proposed system is a novel approach to report generation, providing an efficient and reliable alternative to the traditional approach. The web application’s user-friendly interface and the ability to track report status make it a valuable tool for medical professionals and patients alike.