Medical Image Analysis (IMAG/e)
Main research interest (DSC/e related)
The Medical Image Analysis group (IMAG/e) at TU/e concentrates on automatic analysis methods of medical images to support clinicians in diagnosis, prognosis and treatment. The focus is both on methodological development and clinical application. Methods are generally based on features or biomarkers derived from large sets of medical data or on models verified on extensive clinical datasets. With the rapidly increasing amount of medical data used in practice, the trend is towards learning-based techniques to improve the accuracy and robustness of classification, segmentation and detection tasks as well as towards finding novel biomarkers of disease from the wealth of data available.
In pathology, digitization of stained tissue slides is becoming commonplace. This allows enlisting image analysis to support the pathologist with additional information and to relieve him/her of tedious tasks. Visual analysis of slides is a time-consuming and subjective process, with large variability between observers. Image analysis has the potential to substantially improve the process. We have developed techniques to automatically determine the characteristic features of cancer tissues that are used to grade cases and to subsequently select treatment. We have shown that the automatically estimated features have prognostic value similar to human-defined features. Recent results include deep learning approaches to feature estimation and automatic prognosis that show a performance approaching that of pathologists. We are currently discussing with industrial partners how to incorporate these techniques in the clinical workflow of the pathologist
- DLMedIA is a consortium in which the TU/e collaborates with Radboud UMC, UMC Utrecht, Erasmus MC, UvA as well as clinicians and seven companies.
Goal is to advance the clinical application of medical image analysis techniques based on deep learning.
- RetinaCheck is a Sino-Dutch research and screening program for early detection of diabetic retinopathy through automatic image analysis of retinal fundus images.
Aim is to screen a large population in China, where diabetes has become epidemic. The analysis is based on brain-inspired geometric methods, and new deep learning techniques.