Histology Texture Quantification Using ML Probability Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining cancer progression or recurrence risk after surgery are prone to errors, leading to under- or over-treatment due to inaccuracies in classifying texture features in histology samples.
Innovation Solution
A method and system utilizing a trained machine learning model to quantify texture features in histological samples by dividing digital images into sub-areas, predicting biological features, forming a data matrix, and applying image analysis to produce texture feature quantification, enabling accurate diagnosis and prognosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional methods are used to determine cancer progression risk, then treatment decisions can be made, but errors in classification lead to under- or over-treatment
Solution Approach 1:
The patent replaces traditional manual or simple automated texture analysis methods with a machine learning-based system that processes histology images. The ML model learns complex patterns from training data and provides more reliable cancer progression risk classification, directly addressing the accuracy-precision contradiction by using intelligent algorithms instead of conventional measurement approaches.
Solution Approach 2:
The system performs preliminary training of the machine learning model using labeled histology images and corresponding clinical outcomes before actual diagnosis. This preliminary action allows the model to learn optimal feature representations and classification boundaries, improving both reliability and precision when making actual cancer risk assessments.
2Measurement precision
If wet lab procedures are used to analyze biological features, then accurate diagnosis can be achieved, but the process becomes costly and time-consuming
Solution Approach 1:
The patent creates a digital copy of the histology analysis process by training a machine learning model to replicate the functionality of wet lab procedures. The model learns to predict biological features (such as gene expression patterns or protein presence) directly from digital images, providing accurate diagnosis without requiring actual wet lab experiments, thereby eliminating time delays and reducing costs.
Solution Approach 2:
The system substitutes physical wet lab procedures with computational image analysis. Instead of performing biochemical assays or molecular tests in the lab, the machine learning model analyzes digital histology images to infer biological features, achieving similar diagnostic accuracy while dramatically reducing time and cost.
3Productivity
If simple texture analysis is used, then the process is fast and cheap, but hidden textures and spatial variability cannot be identified
Solution Approach 1:
The system performs preliminary training with diverse histology images to teach the ML model to recognize subtle patterns and spatial relationships. This preliminary learning phase enables the model to automatically detect hidden textures and quantify spatial variability during actual analysis, maintaining speed and cost efficiency while avoiding information loss.
Solution Approach 2:
The patent transforms the analysis by changing from simple texture metrics to complex, multi-dimensional feature representations learned by the machine learning model. The model extracts and analyzes multiple parameters simultaneously, including spatial relationships, texture patterns, and biological feature predictions, thereby capturing hidden information without sacrificing processing efficiency.
Data Source
AI summary
There is provided a method for quantifying texture features in histological sample from a tumor sample, comprising:receiving a digital image of the histological sample, thendividing the digital image into a plurality of sub-areas, thenusing a trained machine learning model to predict a presence of at least one biological feature for each of the sub-areas, where a probability for the presence of the biological feature is represented by a value, thenforming a data matrix by arranging the values for the probabilities of the biological features in the same way as the sub-areas are arranged in relation to the digital image, thenapplying image analysis to the data matrix for a set of texture features, to produce a quantification of at least one texture feature.