Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System

Our research focuses on developing a novel platform, which includes integrate pathology information system, image management system, and artificial intelligence to advanced modern pathology. The question NovinoAI is trying to answer is whether such platform can be developed, used in pathology practice, research, and education. Technologies include computer engineering, digital scanning, computational pathology, and AI model training. One of the most important challenges is the availability of a platform, which allows AI to be integrated and to perform tasks in a seamless manner. We demonstrate that integrated platform can be developed and used in pathology practice, research, and education with desirable accuracy. We are continuing to develop more AI models, including foundation models for future research.

[Narrator] To begin, open a web browser and navigate to the prod FlexLIS website. Enter the pre-registered user email and click continue to proceed. An email will be sent to the registered address. Go to the email inbox and click on the provided sign-in link to access the accession list window. Now turn on the power of the digital scanner and load the previously stained glass slide into the scanner. Press the start button to begin scanning the glass slide. The scanned images will automatically get uploaded to the system via direct interface, and will appear arranged in case format, similar to a pathologist slide tray. In the accession list window, view the list of assigned cases and corresponding slide images in the image review window. In the accession list window, click the down arrow next to the case labeled PN 2404 to reveal stains ordered by staff and AI, such as pin four. A green dot on the microscope icon indicates, AI analysis is complete. A yellow dot indicates ASAP detected by AI on this slide. The reviewed ready total values show the image category counts. An exclamation mark after the image number signals quality issues such as missing or blurry images. Next, click the user icon next to case PN 2404 to open the pathologist review window and observe the specimen summary table, listing all specimens alphabetically. View AI generated results for each specimen, including diagnosis such as benign HG pin, ASAP, or cancer, along with descriptions like Gleason scores, tumor length, and grade group. Then examine the color coded diagnosis categories, showing cancer in red, ASAP in orange, HG pin in green, and benign in black. Now click the microscope icon to open the image review window and view the corresponding slide image. Examine the AI annotations with red highlights marking cancer, and green highlights showing Gleason four. Click C to view image C, then click D to view image D, and so on. Finally, click the AI feedback icon to view detailed AI analysis output. The AI model for prostate cancer detection achieved an accuracy of 97% with a precision of 96%, sensitivity of 98%, specificity of 96%, and F1 score of 97%. The AI assisted histopathology system accurately annotated Gleason three, Gleason four, and Gleason five regions on prostate biopsy slides, using a color coded mask overlay. The AI system, automatically, extracted key pathological metrics, including tumor length, percentage, Gleason scores, and grade group from the prostate biopsy. The AI-based urine cytology model automatically identified diagnostically significant cells, including high grade urothelial carcinoma, atypical cells, and suspicious cells, each outlined with color coded boxes.