Breast Cancer Histologic Grading With Multi-Model Slide Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The inherent subjectivity in manual histologic grading of breast cancer using the Nottingham grading system leads to inter-pathologist variability, limiting its prognostic utility, especially for grade 2 tumors, and existing AI systems often focus on only one or two histologic features, resulting in inaccurate predicted grades.
Innovation Solution
A deep learning system employing multiple machine learning processes for mitotic count, nuclear pleomorphism, and tubule formation is used to generate a combined histologic grade, with each process occurring at a patch level and combining scores to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual histologic grading is used, then pathologist expertise is required, but inter-pathologist variability occurs leading to reduced reliability
Solution Approach 1:
The patent replaces manual pathologist grading with an automated machine learning system that uses deep learning models to objectively assess histologic features. The system processes whole slide images through multiple neural networks that independently evaluate mitotic count, nuclear pleomorphism, and tubule formation, then combines these assessments to generate a unified histologic grade, eliminating human subjectivity and inter-pathologist variability.
Solution Approach 2:
The patent divides the histologic grading task into separate independent modules, with each deep learning model specialized in assessing a specific feature (mitotic count, nuclear pleomorphism, tubule formation). This segmentation allows each component to be optimized independently while maintaining overall system coherence through the integration of individual feature assessments into a comprehensive grading output.
2Measurement precision
If existing AI systems assess only one or two histologic features, then computational simplicity is maintained, but grading accuracy deteriorates
Solution Approach 1:
The patent merges multiple deep learning models that assess different histologic features into a unified grading system. Each model independently evaluates a specific feature (mitotic count, nuclear pleomorphism, tubule formation), and the system integrates these individual assessments through a combination mechanism that produces a comprehensive predicted histologic grade, achieving high accuracy by considering all three Nottingham grading components simultaneously.
3Adaptability or versatility
If grade 2 tumors are included in intermediate risk group, then classification simplicity is maintained, but clinical value is reduced due to inclusion of low and high grade tumors
Solution Approach 1:
The patent employs continuous scoring parameters for each histologic feature rather than discrete categorical grades. The deep learning models output continuous predictions for mitotic count, nuclear pleomorphism, and tubule formation, which are then combined to produce a continuous overall grade prediction. This continuous parameter approach enables more precise risk stratification that can distinguish between low-risk, intermediate-risk, and high-risk tumors more accurately than traditional discrete grading systems.
Data Source
AI summary
A machine learning framework for breast cancer histologic grading is described herein. In an example, a method involves accessing a whole slide image of a specimen. The image is processed using a first, second, third, and fourth machine learning process. A first output of the first machine learning process indicates portions of the image predicted to depict tumor cells. A second output of the second machine learning process corresponds to a mitotic count predicted score for a mitotic count depicted in the image, a third output of the third machine learning process corresponds to a nuclear pleomorphism predicted score for nuclear pleomorphism depicted in the image, and a fourth output of the fourth machine learning process corresponds to a tubule formation predicted score for tubule formation depicted in the image. A combined score of a predicted histologic grade of a disease in the image is generated based on the outputs.