Biopsy Slide AI Segmentation for Consistent Mitotic Cell Counting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining mitotic cell density in animal tissues for cancer grading are tedious, time-consuming, and prone to interobserver variation due to subjective observations and poor image resolution, leading to inaccurate and inefficient cancer assessment.
Innovation Solution
Utilizing machine learning models, such as convolutional neural networks (CNNs) and U-Nets, to automate the process of identifying mitotic cells in digital biopsy images, including tissue detection, segmentation, and mitotic figure counting, with techniques like KD-trees and skip-connections to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of medical images is used to identify mitotic cells, then flexibility and adaptability are maintained, but the process becomes tedious, time-consuming, and prone to interobserver variation
Solution Approach 1:
The patent replaces the manual mechanical review process with an automated machine learning system that uses deep neural networks to identify and count mitotic cells in histology images, eliminating the time-consuming manual observation while maintaining or improving identification accuracy
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the raw medical images and the final diagnostic conclusion, where the model processes images through multiple layers of feature extraction and classification to provide consistent, objective mitotic cell counts without direct human observation
2Measurement precision
If manual observation of mitotic cell density is performed, then clinical judgment can be applied, but the area of concern is much smaller than the overall image making observation difficult
Solution Approach 1:
The patent automatically segments the large histology image into multiple smaller regions or tiles, allowing the machine learning model to focus on specific areas containing mitotic cells while maintaining the ability to aggregate results across the entire image, making small areas of concern detectable within large images
Solution Approach 2:
The patent transforms the 2D spatial problem of finding small mitotic areas in large images by adding computational dimensions through multi-scale analysis and feature space transformations, allowing detection across different resolution levels and aggregating evidence from multiple scales
3Reliability
If manual mitotic counting is performed by trained observers, then clinical context can be considered, but poor image resolution provides an additional complication leading to inconsistent conclusions
Solution Approach 1:
The patent replaces variable human observation with a standardized machine learning system that applies consistent algorithms and criteria across all images, eliminating interobserver variation while maintaining the ability to handle complex clinical scenarios through trained model architectures
Data Source
AI summary
In one embodiment, the disclosure provides a method comprising: receiving an image of a biopsy slide; executing first programmed instructions to input the image to a first trained machine learning model programmed as a cancer classifier, thereby outputting a classification; determining, based on the classification, that the image should be further processed; executing a tissue detect ion algorithm on the image of the biopsy slide, thereby outputting a first set of coordinates; executing second programmed instructions to input the image of the biopsy slide and the first set of coordinates to a second trained machine learning model programmed for image segmentation, thereby outputting a binary pixel mask; executing third programmed instructions to generate an enhanced pixel mask using the binary pixel mask; and digitally storing, in computer memory, a second set of coordinates indicated by the enhanced pixel mask, in one embodiment, a 10-high-power field can be generated and digitally stored.