Usecase UMC Utrecht
Challenge
The early diagnosis of a recurrent tumor for pancreas cancer is crucial. However, it is difficult to distinguish scars resulting from surgery from a recurrence. Datacation was commissioned to develop an AI application that can distinguish these scars on CT scans from recurrence, to support doctors in diagnosing.
Process
The domain knowledge of the medical specialists of the UMC Utrecht has been supplemented with the technical expertise of Datacation. With the input from the specialists, Datacation was able to develop an algorithm that segmented the pancreas using self-supervised learning. This approach is particularly interesting since medical data is limited and self-supervised learning offers the solution. We trained a U-net using self-supervised learning based on the pixels on the CT scans, which segments the pancreas.
Solution
Our AI application supports the medical specialists of the UMC Utrecht in the early detection of pancreatic recurrences. In three months, team Datacation has developed an algorithm that approaches the Dice Score, which is the baseline/benchmark of the literature. This makes it a very valuable addition for the detection of recurrences. Therefore, we are proud to be able to contribute to innovation within healthcare with our algorithm. The next step is to develop a classification algorithm that determines whether or not the tumor has recurred.