CT Radiomic Feature Extraction for NSCLC Chemotherapy Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current treatments for non-squamous non-small cell lung cancer (NSCLC) using pemetrexed-based chemotherapy lack predictive markers, resulting in approximately 24-31% response rate as initial treatment and 11.5% in second-line settings, leading to ineffective use of expensive and side-effect-prone therapies in non-responding patients.
Innovation Solution
A computerized method using machine learning classifiers and radiomic features extracted from computed tomography (CT) images of tumoral and peritumoral regions to predict response to pemetrexed chemotherapy, identifying discriminative texture and shape features that differentiate responders from non-responders.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pemetrexed-based chemotherapy is administered to all NSCLC patients, then treatment coverage is maximized, but resource waste and side effects increase for non-responding patients
Solution Approach 1:
The patent extracts and analyzes radiomic features from baseline CT images before chemotherapy administration to predict patient response. This preliminary analysis enables identification of likely responders and non-responders before treatment begins, allowing clinicians to avoid administering pemetrexed to patients unlikely to benefit, thereby preventing resource waste and unnecessary side effects while ensuring treatment is directed toward those most likely to respond
Solution Approach 2:
The patent introduces radiomic features extracted from CT images as an intermediary biomarker to predict chemotherapy response. These features serve as a mediator between the patient's tumor characteristics and the chemotherapy outcome, providing an objective basis for treatment selection without requiring trial-and-error approaches
2Adaptability or versatility
If pemetrexed chemotherapy is administered to all NSCLC patients, then treatment accessibility is maximized, but harmful side effects increase for non-responding patients
Solution Approach 1:
The patent performs preliminary prediction of chemotherapy response using baseline CT radiomic features before treatment initiation. This allows clinicians to identify patients unlikely to respond and avoid subjecting them to harmful side effects, while maintaining accessibility for those predicted to benefit, thus adapting treatment allocation to individual patient characteristics
3Measurement precision
If comprehensive radiomic feature extraction and machine learning analysis are implemented, then prediction accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the complex prediction task into distinct components: (1) extraction of specific radiomic features from CT images, (2) selection of discriminative features using statistical tests, and (3) classification using machine learning models. This segmentation allows systematic processing of complex data while maintaining manageable system architecture and enabling stepwise optimization of each component
Solution Approach 2:
The patent extracts only the most discriminative radiomic features from the full set of possible image characteristics using statistical comparison between responder and non-responder groups. This extraction process eliminates redundant features and retains only those with predictive value, reducing input complexity to the machine learning model while preserving prediction accuracy
Data Source
AI summary
Methods, apparatus, and other embodiments predict response to pemetrexed based chemotherapy. One example apparatus includes an image acquisition circuit that acquires a radiological image of a region of tissue demonstrating NSCLC that includes a region of interest (ROI) defining a tumoral volume, a peritumoral volume definition circuit that defines a peritumoral volume based on the boundary of the ROI and a distance, a feature extraction circuit that extracts a set of discriminative tumoral features from the tumoral volume, and a set of discriminative peritumoral features from the peritumoral volume, and a classification circuit that classifies the ROI as a responder or a non-responder using a machine learning classifier based, at least in part, on the set of discriminative tumoral features and the set of discriminative peritumoral features.