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

An at least partially computer implemented method classifies a human cutaneous tissue specimen by obtaining a computer readable image of the human tissue sample and preprocessing the image. The method applies a trained deep learning model to label each of a plurality of image pixels with at least one probability representing a particular diagnosis, thereby obtaining a labeled plurality of image pixels.

After obtaining the labeled plurality of image pixels, the method applies 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. The specimen level diagnosis comprises at least one of basal cell carcinoma, dermal nevus, or seborrheic keratosis, and the method outputs the specimen level diagnosis as the final diagnostic result for the specimen.

In a computer system implementation, at least one electronic processor executes instructions to perform operations corresponding to obtaining the computer readable image, preprocessing the image, applying the trained deep learning model to label each of the plurality of image pixels with diagnosis probabilities, and applying the trained discriminative classifier to contiguous regions defined by the labeled pixels. Optional workflow features described in the provided text include tissue-region detection and isolation, heat map generation for the labeled pixels, and confidence level and predetermined threshold logic tied to output to an electronic medical record system.

Claims Coverage

The independent claim coverage centers on a pipeline with three main inventive features: pixel-level labeling using a trained deep learning model, specimen-level diagnosis using a trained discriminative classifier applied to contiguous pixel regions, and output of a specimen-level diagnosis limited to basal cell carcinoma, dermal nevus, and/or seborrheic keratosis.

The independent claims require trained deep learning pixel labeling with diagnosis probabilities, a trained discriminative classifier that operates on contiguous pixel regions defined by the labeled pixels to produce a specimen-level diagnosis, and output of a specimen-level diagnosis comprising at least one of basal cell carcinoma, dermal nevus, or seborrheic keratosis.

Stated Advantages

Documented Applications