Medical Report Text Model Adaptation via Similar Image Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for creating medical reports from medical images are inefficient, as they often rely on modifying past reports of the same patient, which may not be relevant due to disease progression, and result in similar reports for similar images, requiring significant time and effort from medical staff.
Innovation Solution
A report creation support system that retrieves similar case images from a database and generates reports using changeable text models, allowing for automatic creation of reports by analyzing features of the diagnosis target image and replacing words/phrases in the text model with descriptive terms, reducing the need for manual input from doctors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If past reports of the same patient are used for creating new reports, then report creation time is reduced, but report accuracy deteriorates due to disease progression making past reports irrelevant
Solution Approach 1:
The system segments the report creation process into two parts: automatically retrieving and adapting text from similar cases, and allowing manual modification by doctors. This segmentation enables efficient use of past reports while maintaining accuracy through human oversight.
Solution Approach 2:
The system creates copies of text from similar past cases and adapts them to the current case. Instead of using past reports directly, it copies relevant text segments and modifies them based on current image features, reducing time while maintaining accuracy.
2Adaptability or versatility
If free-form text entry is used for medical reports, then report flexibility is improved, but work efficiency deteriorates due to time-consuming manual creation
Solution Approach 1:
The system performs preliminary action by pre-storing structured text templates from past reports, organized by anatomical regions and lesion types. When creating a new report, these pre-prepared templates are automatically retrieved and adapted, eliminating the need for manual free-form text entry while maintaining flexibility.
Solution Approach 2:
The system changes parameters by transforming unstructured free-form text into structured text templates with defined parameters (anatomical region, lesion type, characteristics). This structuring enables automatic retrieval and adaptation while preserving the flexibility needed for diverse medical reporting scenarios.
3Speed
If similar case reports are retrieved and used directly, then report creation speed is improved, but report precision deteriorates due to lack of customization for the specific diagnosis target
Solution Approach 1:
The system introduces an intermediary process between retrieving similar case reports and finalizing the current report. This intermediary step automatically adapts the retrieved text by replacing placeholder parameters (e.g., {{anatomical_region}}, {{lesion_characteristic}}) with values specific to the current diagnosis target, ensuring both speed and precision.
Solution Approach 2:
The system makes the report template dynamic by allowing automatic adaptation of retrieved text based on current case parameters. The template structure remains fixed for consistency, but the content is dynamically customized to match the specific diagnosis target, balancing speed and precision.
Data Source
AI summary
Case images and report text models, each of which is a text model derived from a report text of each of the corresponding case images by making at least certain words/phrases within the report text changeable, are stored in association with each other in a case report storage unit. A case image which is similar to a diagnosis target image is retrieved from the case images stored in the case report storage unit by a similar image retrieval unit. Input of a word/phrase corresponding to the diagnosis target image is accepted in a changeable word/phrase section of the report text model by a report creation unit, thereby a report text of the diagnosis target image is created.