The Swedish medical image company ContextVision has filed a patent for a new method using high-precision deep learning with digital pathology.

The market for their method is estimated to be worth $157 million by 2018. This method allows the company to develop new, unique decision support tools based on true data instead of subjective evaluations.

“The method we have invented enables the development of new products with much higher accuracy and precision, and hence ContextVision will have a unique competitive edge,” says Anita Tollstadius, CEO.

Histopathology samples are currently stained with hematoxylin-eosin in clinical routines, creating images that are challenging to interpret and therefore subjective in evaluation by different pathologists, resulting in significant variation. The inter-observer variability in Gleason grading of prostate biopsies is especially prominent, with disagreement seen in up to 46% of the cases. There is therefore a great need for improved diagnostic methods, as inadequate diagnosis results in poor treatment.

The patent application covers a method that uses super-specific staining with immunofluorescence. This enables training of the deep learning algorithms to use objective true data instead of subjective evaluations, but the developed analysis will be applied to routine HE stained samples.

“The new invention has broad implications for ContextVision as the method can be used for many different cancer types, even if we initially focus on prostate cancer,” says Tollstadius.

Photographer: Alison de Mars von Blixen