AI for Oncology Lab @ Netherlands Cancer Institute
The mission of the AI for Oncology Lab is to develop artificial intelligence innovations for the improvement of cancer diagnostics and therapy.
Retrospective k-space Subsampling schemes For Deep MRI Reconstruction
In our new publication, we investigate and compare various retrospective k-space subsampling patterns and their effect on the quality of DL-based reconstructions. Our findings suggest that non-rectilinear and non-Cartesian subsampling patterns may be more suitable for DL-based reconstructions.
New A100 80GB server installed
Another compute node has been installed in the AI for Oncology Cluster kosmos. The server, nicknamed euctemon, consists of 8xA100 80G, dual CPU and 1TB of memory. Euctemon joins the slurm cluster which now consists out of 16xA100 80GB, 16xA6000 48GB and 4x RTX2080Ti, and 1 PB NAS.
Application of Deep Learning in Breast Cancer Imaging
Luuk Balkenende has published the first paper of his PhD in Seminars in Nuclear Medicine on "Applications of Deep Learning in Breast Cancer Imaging" where he reviews the current usages of deep learning for mammography, ultrasound and breast MRI.
Recurrent Variational Network presented at CVPR
Our paper "Recurrent Variational Network: A Deep Learning Inverse Problem Solver applied to the task of Accelerated MRI Reconstruction" has been accepted for publication at CVPR 2022! CVPR is the top ranked Computer Science conference with leading h5-index and impact score! Our work proposes a novel DL Inverse Problem solver, the RecurrentVarNet, employed and evaluated in the essential task of Accelerated MRI Reconstruction achieving SOTA results!
Open-source Deep MRI Reconstruction software published
Our open-source Deep Image REConstruction Toolkit (DIRECT) has been accepted for publication in the Journal of Open Source Software (JOSS)! DIRECT stores multiple DL model baselines such as the Recurrent Variational Network, implemented in PyTorch for end-to-end Accelerated MRI Reconstruction tasks and allows for use with datasets such as the fastMRI and Calgary Campinas Datasets!
DeepSMILE published in Medical Image Analysis
In this work we use whole-slide image (WSI) compression and multiple instance learning to predict homologous recombination deficiency and microsatellite instability from breast cancer and colorectal cancer WSIs. Both these labels are closely related to a patient’s response to immune- and targeted therapies.
The research projects of the AI for Oncology Lab
Breast MRI is the most sensitive technique for breast cancer detection available. However, the costs and availability of MRI scanners limit its use. We develop algorithms for the automatic interpretation and acceleration of breast MRI.
Medical imaging is the cornerstone of modern medicine. We build deep learning algorithms to reconstruct the measured machine data to an image of the patient anatomy.
Immunotherapy outcome prediction
Immunotherapy is a systemic cancer treatment that exploits the power of the body’s immune system to fight cancer cells by boosting the immune response. We develop algorithms to predict immunotherapy outcomes.
Novel AI methodologies for Oncology
AI enables new diagnostic and treatment paradigms. However, its application to oncology brings many questions and challenges. Inspired by the oncological application, we research and develop new AI methodologies.
The people working at the AI for Oncology Lab