Biological Image Tiling for Accurate Drug Response Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional patient response prediction techniques for pharmaceutical drugs are inaccurate and computationally inefficient due to the large size of biological images, requiring billions of parameters and failing to extract sufficient features from medical images, leading to prediction accuracy ranging from 20%-45%.
Innovation Solution
Implementing image processing and machine learning techniques to process biological images by generating image tiles, preprocessing them to identify discrete tissue components, and using an artificial neural network (ANN) to predict patient responses, such as efficacy or safety responses to drugs, by aggregating values from these components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning techniques are used to process large biological images, then comprehensive feature extraction is attempted, but computational requirements become excessively high (billions of parameters) and prediction accuracy remains low (20%-45%)
Solution Approach 1:
The patent divides the large biological image into multiple smaller image tiles, which are then processed independently. This segmentation reduces the computational burden on each processing unit while maintaining the ability to capture comprehensive features across the entire image through aggregation of tile-level predictions.
Solution Approach 2:
The patent extracts and focuses on specific discrete tissue components within the image tiles rather than processing all pixels uniformly. By identifying and extracting relevant tissue components, the system reduces computational requirements while improving prediction accuracy by concentrating on diagnostically significant features.
2Measurement precision
If traditional techniques process entire biological images, then all regions are analyzed, but computational efficiency decreases and key diagnostic features may be lost in the large data volume
Solution Approach 1:
The patent applies different processing strategies to different regions of the image based on their diagnostic importance. By identifying discrete tissue components and focusing computational resources on these localized regions of interest, the system achieves higher feature extraction accuracy without sacrificing overall computational efficiency.
Solution Approach 2:
The patent extracts specific discrete tissue components from the larger image context and processes them separately. This extraction allows the system to focus computational effort on the most diagnostically relevant features while maintaining efficiency by not processing all image data with the same level of detail.
3Reliability
If large biological images are processed with high computational power, then more parameters can be used, but the system becomes less efficient and more resource-intensive without proportionally improving accuracy
Solution Approach 1:
The patent segments the large-scale prediction problem into smaller, independent tile-level predictions. This segmentation allows the system to achieve reliable overall predictions through aggregation of multiple smaller predictions, reducing the computational power required compared to processing the entire image as a single unit.
Solution Approach 2:
The patent processes multiple smaller copies (tiles) of the original image independently and aggregates the results. This copying approach allows the system to achieve reliable predictions through ensemble methods while reducing the computational power consumption associated with processing the full-resolution image directly.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In an aspect, a data processing system includes a computer-readable memory comprising computer-executable instructions, and at least one processor configured to execute executable logic including at least one artificial neural network trained to predict one or more responses to a chemical substance by identifying one or more discrete biological tissue components in a biological image. When the at least one processor is executing the computer-executable instructions, the at least one processor is configured to carry out operations including: receiving spatially arranged image data representing a biological image of a patient; generating spatially arranged image tile data representing a plurality of image tiles; processing the spatially arranged image tile data through one or more data structures storing one or more portions of executable logic included in the artificial neural network to predict one or more responses of a patient.