Method, System and Computer Program for Detection of a Disease Information

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention introduces a method and system for generating enhanced training data to improve AI-based disease detection in medical imaging. By splitting and mirroring imaging data from symmetric organs (like the brain's hemispheres or paired organs), the method exponentially increases the training dataset size for deep learning algorithms. This makes it possible to better classify, localize, and characterize diseases (e.g., strokes or large vessel occlusions) using medical images with higher accuracy and reliability. The method also includes techniques for optimal data assembly, channel handling (such as vessel or bone data), and visualization optimization for diagnostic support.

Use CasesContent extracted from patent full text and abstract with AI.

• Automated detection and localization of strokes or vascular occlusions in brain scans (CT, MRI, or CTA)
• Augmenting training data for AI systems in diagnostics for any symmetric organ, such as kidneys, lungs, or breasts
• Supporting radiologists with computer-aided diagnosis tools that provide faster and more reliable suggestions
• Development of AI-based triage or pre-diagnosis systems in hospitals and clinics to detect urgent pathologies in medical images
• Enhancement of medical imaging software platforms to include optimal projection views, improving radiologist efficiency
• AI research and development in medical image classification and localization for asymmetric disease detection

BenefitsContent extracted from patent full text and abstract with AI.

• Greatly increases the amount and diversity of training data from limited patient datasets via data recombination and mirroring, improving AI accuracy
• Improves detection and localization of diseases, especially those with subtle or unilateral (one-sided) manifestations
• Allows for balanced AI model training even when original datasets are biased (e.g., mostly right-sided or left-sided anomalies)
• Reduces manual labeling workload, as the mirroring technique synthetically generates variance in the training data
• Enables better generalization of AI models to new, unseen patients through improved training diversity
• Supports rapid and more informed decisions by radiologists with AI-powered view recommendation and enhanced image analysis
• Can be flexibly applied to various symmetric organs beyond the brain, making the technique broadly useful in medical imaging

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

A method for generating training data (1) for training a deep learning algorithm (6), comprising:- Receiving (RT1) medical imaging data (2) of an examination area of a patient, the examination area of the patient comprising a first and a second part of a symmetric organ,- Splitting (SP1) the medical imaging data (2) along a symmetry plane or symmetry axis into a first dataset (3a) and a second dataset (3b), wherein the first dataset (3a) comprises the medical imaging data (2) of the first part of the organ and the second dataset comprises the medical imaging data (2) of the second part of the organ,- Mirroring (MD1) the second dataset (3b) along the symmetry plane or symmetry axis,- Generating (GT1) the training data (1) by stacking of the first dataset (3a) and the mirrored second dataset (3b*)- Providing (PT1) the training data (1).