Register for free to
view the full innovation profile.
Our solution classifies whole-slide histopathology images into phenotype categories associated with predefined pathologies for e.g. gastrointestinal biopsy reporting.
About
In a histopathology laboratory, each diagnosis usually requires skilled specialists to identify diseased tissues though microscopic examination. This process tends to be time consuming and repetitive. Ou machine learning solution helps with diagnosing patients with various tissue phenotypes and helps histopathologists increase the laboratory's efficiency and diagnosis accuracy. Our solution is based on a state-of-the art machine learning technology: We perform feature extraction by a domain-adapted convolutional neural network in a sliding window scheme and aggregate results over multiple scales up to all whole-slide images of a patient for final prediction. The method can utilize hard negative mining, model ensembles and automatic hyperparameter tuning by a proprietary local-search algorithm to boost performance. Due to the underlying technology, the solution provides explanations for the decision made.
Key Benefits
Our technology brings in many benefits: • First of all, the automated detection of pathologies improves laboratory staff workflow and efficiency as they can either process larger amounts of samples or spend their time on downstream tasks. • Second, the accuracy of pathology detection increases as the process is more standardized and sensitive to more nuanced phenomena that might escape the human eye. • Third, automating histopathology image analysis is scalable and can lead to cost-savings as less human labor is used on this task.
Applications
The technology was already applied in the following domains and achieved competitive results in various benchmarks: • Breast cancer detection • Prostate cancer detection • Glioblastoma histopathology slides With some alterations, the technology can be applied in further domains.