A Novel Topic Modeling Framework for Automated Surveying of Computer Vision Research in Prostate Cancer Detection

Retrieving the relevant information or material makes a big difference in the medical field because it is necessary to acquire updates on recent findings and inferences so that the practitioners can understand the state-of-the-art topics. Thus, extracting relevant material automatically from the pool of large databases like Web of Science and IEE Explore on particular issues is important. This work focuses on extracting articles on different topics for prostate cancer detection. Due to a huge number of articles in large databases with many variations in the format, keywords, text in various forms, etc., extracting relevant articles is an open challenge. This observation motivated us to propose new Agnostic-topic modeling for retrieving articles across different categories, namely, Tasks (Image Classification, Image Segmentation, Gleason Grading, Object Detection, Image Enhancement, Motion Tracking, Image Registration, Image Synthesis, and Image Reconstruction), Models (CNN, UNet, GAN, Convolutional ML models, and Uncategorized) and Image Data Type (MRI, CT, Histological Images, Ultrasound Images, Nuclear Imaging, and Others). To address the above open challenge, the proposed work adapted ChatGPT and compared it with well-known models like LDA and BERT to show the robustness of ChatGPT. Furthermore, the extracted information is used to derive new inferences and findings. For example, the trend analysis according to the abovementioned tasks, models, and image types. To the best of our knowledge, this is the first work on retrieving articles automatically for prostate cancer detection.