Deep Learning Liver Pathology Scoring for Continuous NASH Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current manual histological scoring systems for nonalcoholic steatohepatitis (NASH) suffer from high inter- and intra-observer variability and are unsuitable for capturing the continuum of disease severity, leading to challenges in powering clinical trials and assessing treatment response accurately.
Innovation Solution
A machine learning-based approach using deep learning models, specifically deep convolutional neural networks, is developed to assess liver pathology by predicting tissue characteristic categories such as steatosis, lobular inflammation, and fibrosis stage, generating continuous scores that capture disease heterogeneity and treatment effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual histological scoring systems are used, then diagnosis can be made based on expert pathologist review, but inter- and intra-observer variability is high and reproducibility is poor
Solution Approach 1:
The patent replaces manual mechanical scoring by pathologists with an automated image analysis system using machine learning algorithms. The system processes histological images through trained models that objectively quantify fibrosis stages and NASH activity scores, eliminating human observer variability while maintaining diagnostic accuracy.
Solution Approach 2:
The patent creates digital copies of histological images and processes them through trained machine learning models that have been trained on large datasets of annotated images. The models learn to replicate expert pathologist scoring patterns while providing consistent, reproducible results across different users and time points.
2Ease of operation
If ordinal classification systems are used, then disease staging can be simplified, but the continuum of disease severity cannot be captured
Solution Approach 1:
The patent transitions from traditional ordinal scoring (0-4 scale) to a continuous dimensional assessment by generating probability distributions across multiple fibrosis stages. This allows the system to capture the full spectrum of disease severity and heterogeneity while maintaining interpretability through confidence scores and stage probabilities.
Solution Approach 2:
The patent changes the output parameter from discrete ordinal scores to continuous probability values representing the likelihood of each fibrosis stage. This transformation preserves the simplicity of stage classification while adding the dimension of uncertainty quantification and disease heterogeneity capture through continuous scoring.
3Quantity of substance
If placebo treated patients are included in clinical trials, then control group data is obtained, but high rate of histological response creates challenges for powering trials
Solution Approach 1:
The patent implements a feedback mechanism where the machine learning system quantifies treatment response with high precision, enabling better differentiation between placebo and active treatment effects. The system provides detailed feedback on fibrosis stage changes and NASH activity score modifications, allowing for more accurate power calculations and trial design optimization.
Data Source
AI summary
In some aspects, the described systems and methods provide for a method for training a deep learning model to assess liver pathology, including accessing annotated liver pathology images associated with a group of patients in one or more randomized controlled clinical trials of nonalcoholic steatohepatitis therapy, each of the annotated liver pathology images including at least one annotation describing one or more tissue characteristic categories for a portion of the image, and training the deep learning model based on the annotated liver pathology images to predict the tissue characteristic categories, selected from a group comprising steatosis, lobular inflammation, hepatocyte ballooning, and fibrosis stage.