Adaptive Lesion Linking With Voxel Search Across Timepoints

The patent replaces manual mechanical matching with an automated computational system that uses image registration, feature extraction, and machine learning algorithms to link lesions across time points. The system automatically processes medical images, extracts lesion features, and performs matching without human intervention, thereby eliminating the time-consuming and error-prone nature of manual matching while maintaining or improving assessment accuracy.

The system enables self-service by allowing the lesion linking process to be performed automatically by the computational algorithm without requiring manual intervention. The machine learning model learns from training data and autonomously performs lesion matching, feature comparison, and probability calculation, making the system self-sufficient and eliminating dependency on manual expert assessment.

The patent replaces subjective manual matching with an objective computational system that uses standardized image registration algorithms, feature extraction methods, and machine learning models. The system processes images through defined computational steps, applying consistent criteria for lesion identification and matching, thereby eliminating subjectivity and improving reliability of lesion-level assessment.

The system incorporates feedback mechanisms through iterative optimization of the machine learning model. The model learns from training data, receives feedback on matching accuracy, and continuously improves its performance. This feedback loop ensures that the system maintains high reliability and objectivity by constantly refining its lesion matching capabilities based on performance metrics and ground truth data.

The patent applies segmentation by dividing the lesion linking problem into distinct components: image registration, feature extraction, lesion identification, and matching. Each component is handled by specialized algorithms, with segmentation algorithms identifying lesion boundaries at individual time points and subsequent modules linking these segmented lesions across time. This modular segmentation approach enables robust lesion linking by systematically processing each aspect of the problem.

The patent extends the lesion identification problem from a single time point to multiple time points by adding the temporal dimension. The system processes images across different time points, extracts features from each, and links them based on spatial and temporal relationships. This dimensional extension transforms the problem from static lesion detection to dynamic lesion tracking, enabling robust linking across time points through multi-dimensional feature comparison and probability-based matching.