Heike Grabsch

Maastricht UMC

In short, Heike Grabsch's research entails the following: 

Artificial intelligence for high-throughput histological characterisation of the host anti-tumour immune response in regional lymph nodes from oesophageal cancer patients

'Oesophageal cancer (OeC) patients are currently treated based on the disease stage. There is increasing evidence that the host anti-tumour immune response plays a key role in eliminating tumour cells. The local anti-tumour immune response (immune cells infiltrating the primary tumour) depends on the interaction with the host’s immune cells in the tumour draining lymph nodes (LNs). The clinical importance of the LN-based host anti-tumour immune response has not been investigated in detail, mainly due to lack of objective high-throughput LN characterisation methodology. In this project, we aim to develop a Deep Learning (DL)-based approach to automatically find LN containing digital slides, segment LNs and characterise their microarchitecture in OeCpatients from several clinical trials. We expect to (1) expand our basic understanding of the systemic host anti-tumour immune response including changes induced by conventional chemotherapy and (2) to discover and validate clinically useful LN-based biomarker for OeC patients to determine individual patient follow-up strategy or need for additional treatment.'

Brief summary of progress / results 2023

In the second year of this grant, we have made significant progress and are now moving more and more away from the AI-model developmental stage to the application stage. All materials (scanned HE stained histology slides) are available, the high performance computer which we were able to purchase thnks to this grant works very well and will allow us to now let the developped models run through one study set after the next. We are really looking forward to analyse the data and expect new insights into the locoregional immune response with impact on patient management. Also in the second year of this grant, our available datsets remain unique in the world.

Brief summary of progress / results 2022

In this first year, we experienced a significant delay in the ordering, delivery and installation of the custommade high-spec server necessary for the project. However, the waiting time was used to perform and write up two systematic reviews, one with meta-analysis; to establish new collaborations which will significantly enhance the work in this project and work on improving our existing AI-model (J Pathol Inform 2023 Jan 25;14:100192) by using the computational facilities from the Dept of Precision Medicine, University of Maastricht.

This new model performs significantly better when asked to identify digital slides that contain lymph nodes and segment them by combining a patch-level and pixel-level approach. The AI-model to detect germinal centres and sinuses in lymph nodes (code shared with collaborator A Grigordiasis, King's College London) has been tested and inital quality control of germinal center detection performed by H Grabsch and P Canao look very good. We are excited that this code share will allow us to analyse the quantity and quality (shape, size, location) of germinal centers in lymph nodes from patients with oesophagogastric cancer and relate this to clinicopathological data after the summer.

Publicaties

• ScienceDirect.com, February 4, 2024
  Computational methods for metastasis detection in lymph nodes and characterization of the metastasis-free lymph node microarchitecture: A systematic-narrative hybrid review

I