NL
DE
In recent years artificial intelligence (AI) researchers have made great progress in training computers to detect prostate cancer on medical images. Although these AI models are already reaching human-level performance, they are, unlike humans, still unable to effectively combine information from different types of medical images, such as MRI scans (radiology) and tissue microscopy (pathology).
To develop AI models that learn from images from both radiology and pathology, Daan Schouten is researching how to make a digital prostate reconstruction from pathology images. The research group led by Geert Litjens, professor in AI for radiology and pathology images, recently published their results on digital prostate reconstruction in Nature Scientific Reports on January 17th, 2024.
When prostate cancer patients undergo surgery to have their prostate removed, the prostate is cut into many small tissue pieces, which are placed on small glass slides and digitized to be assessed for cancer aggressiveness by the pathologist. For both doctors and AI models it is difficult to directly relate these small tissue pieces to the prostate MRI where the prostate is imaged in its natural shape. Hence, making a reconstruction of the original prostate from all these tissue fragments is essential to be able to combine information from both specialties.
Reconstructing the virtual prostate from digitized tissue fragments is similar to a jigsaw puzzle where all pieces together form an image, but in this case the image is a full slice of prostate tissue. Although computers nowadays are surprisingly good at solving these puzzles, pathology images capture cell-level detail and can easily be 100.000 by 100.000 pixels, being far too big to handle for any computer and making it impossible to solve the puzzle with current algorithms.
With this new research, a smart algorithm has been developed to solve the puzzle and reconstruct the original prostate which was sliced into many small fragments. The researchers validated the algorithm on pathology images from different hospitals, showing that the algorithm is robust for different methods of slicing the prostate in smaller fragments. In the next step, the team will work on creating a large database of combined radiology and pathology images to develop new AI models that can learn from these different types of images. It is expected that this will lead to even better AI models for prostate cancer detection. For now, the algorithm has been made publicly available to be used by researchers around the world.
Read the publication here
Schouten, D., van der Laak, J., van Ginneken, B. et al. Full resolution reconstruction of whole-mount sections from digitized individual tissue fragments. Sci Rep 14, 1497 (2024). https://doi.org/10.1038/s41598-024-52007-5
