Tumor Size and Location Extraction From Radiology Reports
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing NLP systems fail to accurately extract and structure tumor size and location information from unstructured radiology reports, unable to distinguish between prior and current tumors, and are not integrated into EHR systems at scale.
Innovation Solution
A BERT-based NLP model extracts tumor size and location entities from unstructured data, parses dimensions, links entities, and determines anatomical locations using metadata and dictionaries, producing structured data for EHR integration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual extraction of tumor size and location information is used, then data accuracy is maintained, but time consumption and cost increase significantly
Solution Approach 1:
The patent replaces manual mechanical extraction processes with an automated NLP system based on deep neural networks. The system processes unstructured radiology reports and extracts tumor size and location information automatically, eliminating the need for manual review while maintaining high accuracy through trained models and structured output validation.
2Productivity
If existing NLP systems are used, then some entity extraction is achieved, but accuracy in distinguishing prior and current tumors remains insufficient
Solution Approach 1:
The patent introduces an intermediary structured data representation layer between the unstructured radiology reports and the final extraction results. The NLP system first converts unstructured text into structured intermediate representations that explicitly encode temporal relationships and tumor characteristics, enabling more accurate distinction between prior and current tumors before final output generation.
Solution Approach 2:
The patent changes the parameter representation by using structured data fields with specific attributes (such as temporal indicators, tumor status flags, and structured location hierarchies) instead of unstructured text. This parameter transformation enables the system to accurately capture and distinguish temporal relationships and tumor states that existing NLP systems miss.
3Loss of information
If comprehensive tumor information extraction is implemented, then data completeness improves, but system complexity increases
Solution Approach 1:
The patent segments the tumor information extraction task into distinct structured components: tumor size parameters, location coordinates, temporal indicators, and status flags. Each component is extracted and validated independently by specialized NLP sub-routines, then assembled into a complete structured record. This segmentation maintains data completeness while managing system complexity through modular architecture.
4Measurement precision
If NLP models are trained on annotated datasets, then extraction accuracy improves, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-training the NLP models on large annotated datasets of radiology reports before deployment. The models are trained in advance to recognize tumor size patterns, location descriptors, and temporal indicators. This preliminary training enables the system to achieve high extraction accuracy from the start, eliminating the need for iterative retraining and reducing overall time investment.
Data Source
AI summary
A method including receiving an input of unstructured data; extracting at least one size entity and at least one location entity from the unstructured data using a named entity recognition model; parsing dimensions of the at least one size entity; and outputting structured size data and structured location data.