recurrence. The more precisely the heterogeneous tissue can be determined using microscopic diagnostics, the more promising the treatment prospects are - however, the analysis of histopathological images is a lengthy, complex and costly process for which there is currently no standardized procedure. One solution is machine learning, which can be used to determine these images quickly and precisely.
However, in order to train algorithms accordingly, many high-quality data sets are required on the one hand, and on the other, experts have to divide the images into tumor and non-tumor tissue in advance.
Eye tracking makes human analysis behavior visible
A machine learning model developed at the DigiHealth Institute for the diagnosis of glioblastomas can already distinguish tumor tissue from tumor-free tissue with an accuracy of up to 85%. Research professors Prof. Dr. Johannes Schobel and Prof. Dr. Walter Swoboda now want to increase this rate even further as part of the EXPLAINER project - using eye tracking.
"Discussions with neuropathologists have shown that they find it difficult to put into words their decision-making process when analyzing histopathological images - tumor tissue yes or no? - are difficult to put into words. The indicators for the decision - tumor tissue yes or no - can vary depending on the expert," explains Prof. Dr. Johannes Schobel, head of the project. “We now want to use eye tracking methods to make this implicit expert knowledge visible and use it to develop our machine learning models.”
In focus: pupil movements, fixation points and dwell time
This means that as part of a study, experts, including both medical specialists and medical students, are accompanied by eye trackers while viewing images of tumor tissue and tumor-free tissue. In this way, pupil movements can be tracked, fixation points evaluated and the dwell time on image areas calculated. The results of this process are then compared with the results of the existing machine learning model and incorporated into its optimization. By integrating the results into a medical image archiving system (Picture Archiving and Communication System; PACS), a new model can also be trained - in addition to the model for classifying tissue - which is dedicated to marking conspicuous areas, so-called areas of interest.
Interaction between man and machine: added value for everyday clinical practice
Together with practice partner BS Bucher Systemlösungen GmbH & Co. KG, the HNU researchers want to use EXPLAINER to create added value for research, everyday clinical practice and knowledge transfer. “With its innovative machine learning approaches, the project makes a significant contribution to the introduction of standardization in the evaluation of histopathological image data,” says Prof. Dr. Johannes Schobel. “This offers specialists support in their day-to-day diagnostic work, but also benefits medical students in their training.”
About the funding
The EXPLAINER project is being funded by the Federal Ministry of Education and Research (BMBF) for a period of three years as part of the “Research at universities of applied sciences in cooperation with companies (FH-Kooperativ)” funding guideline.
Contact
Prof. Dr. Johannes Schobel, Research professor in the field of Digital Medicine and Care
