Prostate cancer is the most prevalent form of cancer and the second most common cause of death among men in the United States.
Visual Gleason Grading system assessment of histological specimens remains one of the most powerful prognostic predictors in prostate cancer. However, the histological assessment by pathologists of these tissue slides to evaluate the morphological patterns is often time-consuming and suffers from limited reproducibility. We hereby present a deep learning approach for automated Gleason grading of prostate cancer on digital images with Hematoxylin and Eosin (H&E) staining to overcome the above mentioned challenges and deliver reproducible results in a more objective manner.
Our Machine Based Scoring Algorithms
Our Solutions To Resolve The Challenges Appearing From Gleason Grading
Fully Automated Solution
End to end solution with robust and efficient algorithm modules
- Intelligent segementation module that works on human perceptible color spaces to detect cell nuclei based on recognizable patterns like area, shape, intensity etc.
- Automatic detection of glandular lumens based on the clustering of identified cell nuclei and other features.
- Robust feature extraction module to extract structural, morphometric, texture, nucleocytoplasmic ratio and color features for detected cell nuclei and identified glandular regions.
ANN (Artificial Neural Network) Based Classifier
- Feature fusion and feature ranking techniques for representation to the Neural network-based classifier.
- The classifier is trained to distinguish between moderately and poorly differentiated glands.
- Object level tumor grading is done using feature characteristics for malignant and benign cell nuclei like mean intensity, area, standard deviation of intensity etc.
- Easily retrainable machine learning system.
- High classification accuracy.