Linda Douw

Amsterdam UMC

In short, Linda Douw's research entails the following:

Predicting functional and cognitive decline after glioma surgery (PREDICT)

The PREDICT project aims to forecast postoperative decline in glioma patients using clinical and advanced ‘connectome’ or brain network-based predictors. We will first train a classifier for decline using only demographic and generic computational modeling, and then personalize this model with advanced connectome data. Finally, we will validate the model in a new patient cohort.

This project pivotally uses concepts from "cancer neuroscience," exploring the interaction between tumors and the brain in order to forecast relevant clinical outcomes. It uniquely combines generative computational modeling with machine learning to potentially improve oncological care. Ultimately, the project aims at reducing uncertainty for patients facing surgery and their healthcare professionals, and potentially enabling personalized (p)rehabilitation strategies to enhance brain resilience in those at risk of decline.

The PREDICT team consists of a neurosurgeon with a keen interest in optimizing patient outcomes through advanced analyses (prof. Philip De Witt Hamer), a physicist/MD who has ample expertise in computational modeling (dr. Prejaas Tewarie), a technical physician with lots of AI know-how (dr. Roelant Eijgelaar), and a neuropsychologist/multiscale network neuroscientist with experience in combining imaging and cognition in these patients (dr. Linda Douw).

Short summary of progress/results April 2026

The PREDICT team has successfully completed its first study with a manuscript currently preprinted and submitted for publication. We included 552 patients and built a state-of-the-art machine-learning model to predict one-year postoperative functional status in patients with contrast-enhancing glioma at the preoperative stage. This is inherently challenging as histopathological reports and postoperative treatment information are not yet available. Results showed that most MRI-based predictors did not improve predictions as the best-performing model included three predictors: age at diagnosis, contrast-enhancing volume, preoperative KPS. Bootstrapped areas under the curves were 0.77 (95% confidence interval 0.70–0.84) for mortality, 0.64 (0.52–0.77) for functional dependence, and 0.71 (0.63–0.79) for functional independence. Mondrian conformal prediction provided reliable predictions for 18% patients, moderate uncertainty for 57%, and identified 25% with genuinely unpredictable outcomes. The model is freely accessible via this web-based app.

We are now advancing to the next phase, focusing on glioma computational modeling. In the next phase, we will focus on using computational modeling to simulate the personalized preoperative brain network models of patients with glioma. This will pave the way for a ‘virtual resection’, a process of simulating neural activity after removing white matter fibers that intersect with the actual surgical resection mask.
Ultimately, such a computational modeling approach will provide additional predictors for the preoperative prediction model.

Short summary of progress/results April 2025

In the first stage of this project, we conducted a preliminary study involving 153 glioma patients with grade 2 and higher. The goal was to develop predictive models for patient survival and functional outcomes, as measured by the Karnofsky performance score, one year after resection. By using a combination of clinical and demographic data, we trained several machine learning models. These models were able to predict outcomes with 69–74% reliability. These initial findings were presented with a scientific poster at the Amsterdam Neuroscience Annual Meeting in December 2024.

In the next phase, we are expanding the study to include 541 patients and adding metrics extracted from patients’ brain scans. This will enable us to develop more accurate machine learning models to predict a key outcome: whether glioma patients are likely to retain or experience a decline in functional abilities one year after their first resection. This will provide crucial insight into whether these patients are able to lead a fulfilling independent life after the full first cycle treatment is over.