Virtual Biopsy Model for Medical Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current histopathological-immunohistochemical and genomic analyses of ex vivo biopsies are invasive, limited by partiality and regional randomness, and struggle to capture the full architectural, morphological, and functional complexity of tissues.
Innovation Solution
A method for generating models to automatically classify medical or veterinary images into classes of interest by processing original images to extract tissue partitions or functions, using virtual contrast media to enhance image analysis, and integrating advanced data selection, training, and validation procedures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If histopathological-immunohistochemical and genomic analyses of ex vivo biopsies are performed, then tissue characteristics can be identified, but the procedures are invasive and limited by partiality and regional randomness
Solution Approach 1:
The patent creates virtual biopsies by generating synthetic images that replicate the appearance and characteristics of real tissue biopsies. These virtual biopsies are derived from medical images (CT, MRI, PET) and contain realistic tissue structures, cellular patterns, and pathological features without requiring actual tissue sampling. The virtual biopsies preserve all diagnostic information while eliminating the need for invasive procedures.
Solution Approach 2:
The patent replaces physical biopsy procedures with computational image processing and generation techniques. Instead of mechanically extracting and analyzing tissue samples through immunohistochemical staining and genomic sequencing, the system uses algorithms to process medical images and generate virtual tissue representations that can be analyzed in silico, substituting wet lab procedures with dry computational methods.
2Loss of information
If real biopsies with contrast media are performed, then detailed tissue characteristics become visible, but the procedure becomes more invasive and complex
Solution Approach 1:
The patent generates virtual biopsies that replicate the contrast-enhanced appearance of real tissue without actually administering contrast media. The virtual biopsies incorporate realistic tissue structures, cellular patterns, and pathological features with enhanced visibility, allowing detailed tissue characterization through computational generation rather than chemical enhancement.
Solution Approach 2:
The patent introduces virtual biopsies as an intermediary between medical imaging and diagnostic analysis. Instead of directly performing invasive biopsies with contrast media, the system first generates virtual representations from non-invasive medical images, then analyzes these virtual biopsies to extract diagnostic information, serving as a non-invasive mediator that preserves all diagnostic capabilities.
3Measurement precision
If advanced image processing is applied to medical images, then tissue characteristics not visible to the naked eye can be predicted, but the computational complexity increases
Solution Approach 1:
The patent creates virtual biopsies that contain enhanced and amplified tissue characteristics derived from medical images. The virtual biopsies preserve and emphasize subtle tissue features, cellular patterns, and pathological markers that are not visible in original medical images, making them detectable through standard analysis methods without requiring complex computational processing of the original images.
Solution Approach 2:
The patent performs preliminary image processing and feature extraction during the virtual biopsy generation phase. Complex computational operations, including image registration, segmentation, and feature enhancement, are executed upfront when creating the virtual biopsies. This preliminary processing prepares the data in advance, allowing subsequent diagnostic analysis to proceed with simpler, more efficient algorithms.
Data Source
AI summary
A method to generate predictive models to automatically classify, medical or veterinary images derived from original images is disclosed. The method is computer-implemented and includes the steps:selection of at least one class of interest wherein the classes of interest characterize predictions;construction of a database containing a plurality of data structures representative of derived images and generated by processing original images in such a way that they can be used for surrogate of real biopsies with real staining media performed by means of medical imaging systems;association to the classes of interest of said data structures; andtraining of at least one model using the data structures on the basis of the differences in the distribution of expression levels of measured characteristics of the derived images, to classify each of the derived images in classes of interest.