23rd January, 2017
HeteroGenius are pleased to announce the release of the Deep Learning add-on for their Medical Image Manager (HG-MIM) web based digital pathology platform. Deep Learning, using convolutional neural networks, has been shown time and time again to outperform more traditional feature based machine learning tissue recognition, often by a large margin. Utilising the power of NVIDIA GPU cards powerful models of tissue appearance may be built based on annotated examples that recognise tissue type in a similar way to how the human brain recognises appearance. Once generated these models may be used for whole slide and sub-region quantitation (including high resolution whole slide overlays), automatic annotation and automatic stereology. Whole slide analysis at high resolution can be performed in minutes even on modest GPU hardware.
Dr Derek Magee HeteroGenius CTO said ‚ÄúThe power of the HeteroGenius deep learning add-on is that models may be built using a relatively small number of annotations taking only minutes to perform ‚Äď rather than the user having to laboriously hand annotate thousands of examples. Novel training data augmentation and colour normalisation functionality allows models to be applied to a wider range of staining variation not seen in the training set and avoids the problem of ‚Äėoverfitting‚Äô normally associated with using smaller amounts of training data in which models work well on the training data, but perform badly on other data.‚ÄĚ
The Deep learning add-on is fully compatible with other elements of the HeteroGenius Medical Image Manager including the MIM-Pro report generation system (for exporting quantification results to Excel, SPSS etc.), and the 3D Pathology Add-on ‚Äď allowing volumetric visualisation export of results as colour-coded volumes, or triangulated surfaces.
For more information: