AI for rapid tumour diagnosis

The microscopic examination of tissue samples and cell smears plays a major role in cancer diagnostics. Pathologists not only differentiate between benign and malignant tumours, but also define tumour type, stage and progression. This is the prerequisite for selecting the optimum therapy. This is where microscopic examination alone reaches its limits. Prof Wolfram Klapper is hoping for progress from digital pathology. The head of the Hämatopathology and Lymph Node Registry Section at the Institute of Pathology at the University Medical Centre Schleswig-Holstein (UKSH), Kiel Campus, is an expert in the diagnosis of malignant lymphomas, i.e. malignant tumours of the lymph nodes or lymphatic tissue. Together with working groups from Stuttgart, Würzburg, Göttingen and Regensburg, the Kiel pathologist launched the project „Föderiertes Lernen in der Lymphompathologie: Infrastruktur, Modelle, Erweiterungsalgorithmen, Detektion von Hochrisikopatienten (FDLP)“ in November.

KI diagnosis at three sites

So far, pathologists have prepared so-called histological tissue sections from tissue samples and examined them under a microscope. In the new project, these tissue sections are stored as digital images on servers using microscope scanners at the three participating pathology sites (Kiel, Würzburg, Stuttgart). The computer science working groups in Regensburg and Göttingen use this image data and the associated molecular analyses to train artificial intelligence (AI) programmes. „A special feature of the project is that the medical data does not leave the protected space of a location“, explains Prof Klapper from the Faculty of Medicine at Kiel University (CAU). In so-called fused learning, machine learning algorithms can work with data without the data having to leave the storage location. This ensures data security on the one hand and increases the amount of data due to the three locations on the other.

Potential of AI for diagnosis

The aims of the project are to test whether AI can be used to make the correct diagnoses and whether AI can correctly predict a specific tumour type with an aggressive course of disease based on image data alone. Specifically, this involves the detection of B-cell lymphomas with Myc translocation, a special genetic alteration that is associated with a poor prognosis. In order to detect this type of tumour, the tumour DNA must be examined in addition to the microscopic impression. This is because lymphomas all look very similar in tissue sections. There are only a few visually detectable differences.

In the long term, the aim is to integrate artificial intelligence into the diagnostic process and thus provide faster and more precise information for therapy planning. To this end, the researchers can draw on a large archive of lymphoma tissue samples that have long been collected in the Hämatopathology Section at the UKSH, a specialist diagnostic centre active throughout Germany. If the project confirms that digital pathology promises progress that goes beyond mere microscopic examination, this could give digitalisation in pathology a boost, hopes Klapper.

Report of the Faculty of Medicine, CAU from 10 January 2023