University of Twente Student Theses


Prediction of Immunotherapy Outcome in Melanoma Patients with Brain Metastases

Wamelink, Ivar J.H.G. (2021) Prediction of Immunotherapy Outcome in Melanoma Patients with Brain Metastases.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Abstract:Immune checkpoint inhibitors (ICI)(immunotherapy) have provided improved response rates and prolonged survival in patients with metastatic cancer. Multiple clinical trials have shown that the use of immunotherapy provides a better treatment outcome than standard therapy. Especially in metastatic melanoma cancer immunotherapy has shown promising results. Despite the approval of certain ICI by the FDA and EMA, there remains limited information for optimal patient and treatment selection. Furthermore, current guidelines state that all melanoma patients with metastases qualify for immunotherapy. However, besides high costs, 12-20% of patients receiving ICI show adverse events. The high incidence of immune-related adverse events, high costs, and lack of evidence stating which patients will benefit from immunotherapy, an accurate, reproducible, and predictive marker is needed that clinicians can use in their clinical workflow. The predictive marker could function as a decisive factor during treatment planning - preventing unnecessary adverse events, save precious resources for healthcare facilities, and increase the quality of life. To create such a predictive marker we developed three AI models for the prediction of patient survival. First, we created an AI model capable of detecting general brain anomalies where the model returns an anomaly heatmap. The results show that the anomaly heatmaps share a significant relation to age, C-reactive protein, and necrotic tissue. The anomaly heatmaps could provide additional information and insights that can be used in clinical and methodological research on melanoma patients with brain metastases treated with immunotherapy. Second, we developed an automatic segmentation model for tumour compartments (enhancing tissue, necrotic tissue, edema). The predicted volume showed a high correlation to the segmentation volume created by expert readers. There was a significant relation between segmentation volume and patient survival. The AI model can be used to decrease the workload of physicians by longitudinally and accurately monitoring the lesion volume. Finally, we used the segmentation model to construct four new predictive features. All features showed mild correlation to patient survival. Even though the predictive features show promising results, they require further development before clinical implementation is possible.
Item Type:Essay (Master)
Nederlands Kanker Instituut, Amsterdam, Netherlands
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page