Granted applications 2020

Artificial Intelligence in Oncology - Supporting scientific research

Granted applications 2020

Karijn Suijkerbuijk

UMC Utrecht

 

In short, Karijn Suijkerbuijk's research entails the following: 

Predicting immunotherapy response in advanced melanoma

'Immunotherapy with checkpoint inhibitors has importantly changed perspectives for many metastatic melanoma patients. Still, the majority of patients do not derive long-term benefit from this treatment. Currently, no single biomarker is available that can reliably predict who will and who will not respond to immunotherapy. If non-response could be predicted, patients could be spared the potential severe side effects of immunotherapy and start a more effective treatment. Aim of this study is to develop machine learning algorithms based on clinical data and histological images from the primary melanomas of 1500 immunotherapy- treated metastatic melanoma patients that can predict response to immunotherapy. The project is a collaboration between the UMC Utrecht (dr. Karijn Suijkerbuijk, dr. Willeke Blokx, prof. Paul van Diest, prof. René Eijkemans), the TU Eindhoven (dr. Mitko Veta, prof. Josien Pluim) and several Dutch melanoma treatment centers.' 

 

Publications

  • A novel dataset for nuclei and tissue segmentation in melanoma with baseline nuclei segmentation and tissue segmentation benchmarks
    • GigaScience, 2025
    • Click here to view the publication

 

 

  • Baseline tumor-infiltrating lymphocyte patterns and response to immune
    checkpoint inhibition in metastatic cutaneous melanoma

 

Brief summary of progress / results December 2023

For some patients with advanced melanoma, immunotherapy extends survival with multiple years or even cures the disease. For more than half of patients, however, the therapy is not effective. Currently, there is no way of predicting who will respond. The aim of this research project is to predict the effect of immunotherapy by applying artificial intelligence (i.e. machine learning) to microscopy imaging of tumor tissue. In the past two and a half years, we have gathered anonymous data from nearly 3225 patients who received immunotherapy for melanoma. This data includes information on the location, thickness, and subtype of melanoma. Using this information, we have developed a prediction model that can predict response to immunotherapy. However, this model, based only on clinical and pathological data, is not yet sufficiently accurate to refrain from treatment based on its prediction. We hope to increase its precision by incorporating microscopy images of the melanoma. 

To achieve this goal, we have collected microscopy images of melanomas from over 1600 patients. Analyzing these images we have shown that the number of white blood cells present in the melanoma tissue, taken before starting therapy, has predictive value and improves the prediction model based on clinical data. In the coming two years we will focus on further improving these predictions by extracting more information from the microscopy images. We will do this by using artificial intelligence to examine the amount, location and relationship between different immune cell types. In addition, we will use deep learning to analyze the entire image. With this next step, we aim to refine the predictions of our final model to make it more reliable. 

Brief summary of progress / results December 2022

In a large portion of patients with advanced melanoma, immunotherapy is able to extend survival with multiple years or even cure the disease in some cases. For more than half of patients, however, the therapy is not effective. Currently, there is no way of predicting who will respond before the treatment is given. The aim of this research project is to predict the effect of immunotherapy by applying artificial intelligence (i.e. machine learning) to microscopy imaging of tumor tissue. In the first 1.5 years of the project, we thus far collected anonymous data of almost 2500 patients about their disease (for example: location on the body, thickness and type). We are investigating how well this information is able to predict the effect of the treatment. We expect to present these results early 2023, at the end of the second year of the project.

Furthermore, we have so far collected more than 600 microscopic images of primary melanomas. We are currently digitizing these images. Based on the currently available data, we have developed a model that is able to detect individual white blood cells within the melanoma slides. Lastly, we have developed a model for predicting response based on the microscopy imaging. In the coming years, we aim to test our models on the available data, improve our method where necessary and, lastly, test it on the full dataset.

Brief summary of progress / results 2021

Immunotherapy works for some patients with metastatic melanoma so well that they survive for years after treatment and may even be able to to cure. However, this treatment works for more than half of the patients not and at the moment we cannot predict this in advance. The purpose of this research is to see if we can use artificial intelligence (so-called machine learning) on data and microscopy images of the melanoma can predict who will or will not benefit from immunotherapy. In the first 9 months of the study, we have had almost 1000 patients who treated with immunotherapy, the (anonymous) data were collected. We hope to expand this to 1500 in the coming months In addition, we are busy with collect the microscopy images. Meanwhile we have the first steps in creating a program that automatically extracts from those microscopy images can recognize the different cells that are present in the melanoma. In the coming year we will continue to collect the data and will we are going to use this data to build an algorithm that measures the effect of immunotherapy before starting treatment.