Deep learning automated dermatopathology

Inventors

Jackson, Brian H. • Stavish, Coleman C.

Assignees

Proscia Inc

Abstract

Techniques for classifying a human cutaneous tissue specimen are presented. The techniques may include obtaining a computer readable image of the human tissue sample and preprocessing the image. The techniques may include applying a trained deep learning model to the image to label each of a plurality of image pixels with at least one probability representing a particular diagnosis, such that a labeled plurality of image pixels is obtained. The techniques can also include applying a trained discriminative classifier to contiguous regions of pixels defined at least in part by the labeled plurality of image pixels to obtain a specimen level diagnosis, where the specimen level diagnosis includes at least one of: basal cell carcinoma, dermal nevus, or seborrheic keratosis. The techniques can include outputting the specimen level diagnosis.

Core Innovation

The disclosed invention relates to an automated dermatopathology workflow that classifies human cutaneous tissue using computer-implemented processing of whole-slide images. A computer readable image of a human cutaneous tissue is obtained, and a trained model labels a plurality of pixels with at least one diagnosis probability to obtain labeled image pixels. The approach includes identifying regions of pixels associated with a particular diagnosis probability above a predefined threshold and using those regions for further analysis.

The invention further obtains an automated diagnosis using a trained classifier applied to at least one contiguous region of pixels defined in part as a portion of the labeled plurality of image pixels. The trained classifier is trained by tissue images and corresponding diagnostic labels, and the automated diagnosis comprises at least one basaloid, squamous, or melanocytic dermatopathological condition. In the disclosed examples, the automated diagnosis includes basal cell carcinoma, dermal nevus, and seborrheic keratosis.

The disclosed workflow also describes preprocessing to detect and isolate tissue-containing image regions, tiling of whole-slide images into image tiles, and pixel-level probabilistic labeling. Contiguous high-probability regions are extracted from the labeled pixels, and feature vectors are generated for these regions for use by a discriminative classifier. The disclosure further describes trained models including a fully convolutional deep neural network and a support vector machine, along with confidence-thresholded reporting and heatmap visualization for diagnosis probabilities, and provides an example computer system architecture for implementation.

Claims Coverage

The independent claims are directed to a computer implemented method and a computer system. Across both independent claims, four main inventive features are consistently covered: obtaining an image of human cutaneous tissue, pixel-level labeling with diagnosis probabilities using a trained model, deriving an automated diagnosis by applying a trained classifier to contiguous labeled pixel regions above a predefined threshold, and outputting the automated dermatopathological diagnosis including basaloid, squamous, or melanocytic conditions.

The claims cover both method and system embodiments that label pixels with diagnosis probabilities using a trained model, then apply a trained classifier to contiguous labeled pixel regions whose diagnosis probability exceeds a predefined threshold. The classifier is trained using tissue images and corresponding diagnostic labels, and the output is an automated dermatopathological diagnosis comprising at least one basaloid, squamous, or melanocytic condition.

Stated Advantages

Documented Applications

No documented applications found