Imaging is a cornerstone of medicine. The number and volume of radiology exams is growing rapidly thereby tremendously increasing the workload of radiologists. Deep learning methods can potentially extract more information from images, more reliably, more accurately, and most notably fully automatically.
The goal of the DLMedIA programme is two fold: 1) Develop a technological software platform for application development and; 2) build specific solutions for personalized and precision medicine. Through four different projects and in collaboration with four research centers and two industrial partners (Philips Healthcare, Pie Medical Imaging) , in our group we focus on early detection and prevention cardiovascular disease.
High dimensional data
Goal: To develop deep learning techniques for quantitative analysis of 4- and 5-dimensional medical images.
Deep generative models
Goal: To incorporate expert knowledge in the form of generative models in deep learning in order to learn more efficiently from less data.
Application: Blood vessel geometry synthesis using generative adversarial networks
Deep transfer learning
Goal: To develop deep transfer learning techniques to effectively analyze heterogeneous medical imaging data with variation in scanners, scan protocols, and patient populations.
Dynamic deep learning
Goal: To develop dynamic learning strategies for deep learning systems in clinical environments.
Prof. dr. Bram van Ginneken, Diagnostic Image Analysis Group, Radboud University Medical Center, Nijmegen
Dr. Clarisa Sanchez, Diagnostic Image Analysis Group, Radboud University Medical Center, Nijmegen
Prof. dr. Max Welling, Amsterdam Machine Learning Lab, Department of Science University of Amsterdam (UvA)
Dr. Marleen de Bruijne, Biomedical Imaging Group Rotterdam, Erasmus Medical Center Rotterdam
Prof. dr. Josien Pluim Medical Image Analysis, Eindhoven University of Technology
Papers in international journals
|1.||Application of speCtraL computed tomogrAphy to impRove specIficity of cardiac compuTed tomographY (CLARITY study): Rationale and Design. BMJ Open, 2019, vol. 9, nr. 3, pp. e025793.|
Papers in conference proceedings
|1.||Graph convolutional networks for coronary artery segmentation in cardiac CT angiography. In: 1st International Workshop on Graph Learning in Medical Image (GLMI 2019), in press, 2019.|
|2.||CNN-based segmentation of the cardiac chambers and great vessels in non-contrast-enhanced cardiac CT. In: Medical Imaging with Deep Learning (MIDL 2019), 2019.|
|3.||Improving myocardium segmentation in cardiac CT angiography using spectral information. In: SPIE Medical Imaging, 2019.|
|4.||Towards increased trustworthiness of deep learning segmentation methods on cardiac MRI. In: SPIE Medical Imaging, 2019.|
|5.||Blood vessel geometry synthesis using generative adversarial networks. In: Medical Imaging with Deep Learning (MIDL 2018), 2018.|
|6.||CNN-based Landmark Detection in Cardiac CTA Scans. In: Medical Imaging with Deep Learning (MIDL 2018), 2018.|
|1.||New dimensions in cardiovascular CT. Utrecht University, The Netherlands, 2019, ISBN: 978-90-393-7092-6.|