Computer-aided diagnosis (CAD) from medical images has been used to help radiologists for many years. But with the use of High Performance Computing and deep learning algorithms for automatic recognition of complicated patterns in magnetic resonance imaging (MRI), computed tomography (CT), or whole-slide histopathology images (WSI), the capabilities of CAD systems have the potential to be significantly improved.

Computers are now, for the first time, matching the performance or even outperforming medical specialists, while also accomplishing the analysis much faster.

Medical Image Analysis (MRI, CT, WSI) are saving lives daily. These techniques are producing an enormous amount of valuable data that needs to be analysed by the physicians that will transform them in diagnosis. But as medical image acquisition methods become more widespread and accessible, the need for radiologists that can interpret an increasing number of images grows as well. For that reason, automated techniques in analysing data are not only very helpful, but are expected to increasingly become the preferred methods to use.

Combining High Performance Computing (HPC) resources with deep learning algorithms could greatly improve the recognition of complicated patterns in MRI, CT, or whole-slide images.  Being able to further develop, optimize, and allow the hospitals and general practitioners to make use of these powerful techniques could have important societal impact by improving the quality of life through more precise and cost effective diagnostics.

The Computational Pathology group of Radboud University Medical Center (Nijmegen, Netherlands) uses High-Performance Computing to enhance their physicians' diagnostics, providing faster diagnosis and saving more lives.

In the image attached, you can see (left side) an example of colon carcinoma histopathology section stained with Hematoxylin and Eosin (H&E).

On the right side, you can see the result of automatic segmentation (coloring) of multiple tissue types using a deep learning algorithm applied to the digitized image on the left. The algorithm can identify up to 14 different tissue types, which are visualized with different colors according to the color-coding reported in the legend.