Radiomics-Based Breast MRI Diagnosis With Automated Tumor Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies require trained radiologists for diagnosis and are limited in their ability to provide accurate, non-invasive prediction of cancer types and prognostic information, particularly for aggressive cancers like triple negative breast cancer, and do not efficiently utilize the vast amount of radiomic features extracted from images.
Innovation Solution
A system utilizing a combination of variational autoencoders and U-Nets for automated tumor segmentation, followed by selective extraction of key radiomic features, and supervised machine learning models for predictive diagnosis and prognosis, including identification of cancer types and gene mutations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radiologists manually analyze medical images for diagnosis, then diagnostic accuracy can be maintained, but the process is time-consuming and requires specialized training
Solution Approach 1:
The system enables automated diagnosis through machine learning models that independently analyze radiomic features without requiring continuous radiologist intervention. The models self-train on extracted features and automatically generate diagnostic predictions, reducing dependency on manual analysis while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical process of manual radiologist analysis with an automated computational system. Machine learning models process radiomic features extracted from medical images, substituting human cognitive processing with algorithmic analysis to reduce diagnosis time while preserving diagnostic precision.
2Loss of information
If all radiomic features are extracted and analyzed, then comprehensive diagnostic information is obtained, but the computational complexity and processing time increase
Solution Approach 1:
The system extracts only the most relevant radiomic features from medical images using selective feature extraction methods. By identifying and isolating key features that contribute most to diagnostic accuracy, the system reduces the total number of features processed while maintaining comprehensive diagnostic information.
Solution Approach 2:
The diagnostic process is segmented into distinct stages: feature extraction, feature selection, model training, and prediction. This segmentation allows the system to process radiomic features in manageable stages, reducing overall computational complexity while ensuring no critical diagnostic information is lost.
3Measurement precision
If invasive biopsy procedures are performed for cancer diagnosis, then definitive diagnosis is obtained, but patient discomfort and procedural risks increase
Solution Approach 1:
The system introduces radiomic feature analysis as an intermediary between non-invasive imaging and invasive biopsy. By extracting and analyzing quantitative features from medical images, the system provides diagnostic information that can reduce or eliminate the need for invasive procedures, thereby maintaining diagnostic accuracy while minimizing patient harm.
Solution Approach 2:
The system creates a virtual representation of tumor characteristics through radiomic features extracted from imaging data. This digital copy contains sufficient diagnostic information to make accurate cancer subtype predictions without requiring physical tissue sampling, thereby avoiding the harms associated with invasive biopsies.
Data Source
AI summary
Systems and methods for automated diagnosis and prognosis support using radiomics in accordance with embodiments of the invention are illustrated. One embodiment includes a method for non-invasively identifying triple negative breast cancer, comprising obtaining a magnetic resonance imaging scan of a patient's breast, generating a tumor segmentation mask for the scan of the patient using a first machine learning model, extracting several radiomic features from the segmented scan of the patient, providing the several radiomic features to a second machine learning model, and obtaining, from the second machine learning model, a likelihood that the patient has triple negative breast cancer. In a further embodiment, the second machine learning model further provides an estimated survival time of the patient.