Data Characteristics
Data in the pharmacovigilance domain for imaging equipment primarily originates from Picture Archiving and Communication Systems (PACS), Radiology Information Systems (RIS), device logs, maintenance records, and patient medical histories. Data update frequency varies by equipment type and usage. High-traffic devices like CT and MRI scanners typically have imaging reports and device logs updated at a minute level, while X-ray machines might update hourly or daily. Document structures often include unstructured free-text reports, semi-structured device parameters and operation records, and structured diagnostic codes (e.g., ICD-10), along with device models and serial numbers. Imaging reports frequently contain extensive medical terminology, abbreviations, and device-specific parameters (e.g., Dose-Area Product, Tesla). The units for these data fields must be precise, such as radiation dose in mGy·cm² and contrast agent volume in ml.
Constraints Imposed by Data Characteristics on Citation and Traceability
Unstructured free text in imaging reports, such as diagnostic conclusions and imaging descriptions, challenges RAG model segmentation and vectorization. This requires more refined text processing strategies to ensure semantic integrity. Semi-structured device logs and maintenance records often have inconsistent fields, making information extraction difficult and impacting traceability accuracy. Structured device parameters and diagnostic codes are easier to process, but their contextual relevance needs extra attention to avoid misinterpretation. High-frequency data sources, such as CT/MRI reports, demand real-time or near real-time index update capabilities to ensure citation timeliness. Additionally, the specialized medical terminology and units in imaging data require the model to accurately identify and retain this professional information during comprehension and citation. This prevents traceability failures due to misunderstanding or loss of critical information, directly affecting the assessment of adverse drug events related to imaging equipment operations.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunk_size | 500-800 characters | Balances the contextual integrity of medical terminology in imaging reports with vectorization processing efficiency, preventing semantic fragmentation. |
overlap_size | 100-150 characters | Ensures sufficient overlap between chunks to capture cross-chunk logical connections, especially for lengthy imaging descriptions. |
top_k | top 5-8 entries | Improves recall, covering more potentially relevant imaging report segments, device logs, and maintenance records. |
embedding_model | text-embedding-ada-002 or bge-large-zh | Selects a model with strong medical semantic understanding capabilities to enhance the vectorization accuracy of specialized terminology. |
parse_timeout_seconds | 600 seconds | Accommodates the parsing time for large imaging reports or device log files, preventing processing interruptions due to timeouts. |
similarity_threshold | 0.75-0.85 | Balances the precision and coverage of recall, filtering out low-relevance citations to improve traceability quality. |
Common Pitfalls
- Key imaging equipment models or serial numbers are missing from question-answering results. This occurs when these fields are truncated or not effectively recognized during text chunking, leading to information loss during vectorization.
- The system frequently displays "parsing failed" or "unsupported file format" errors when processing certain device log files. This is due to the diverse formats of log files, without adaptive parsing for all variations.
- When a user queries the relationship between an adverse drug event and a CT scan, the returned citations do not match the actual imaging report times. This is because of knowledge base index update delays, failing to synchronize the latest imaging report data in a timely manner.
Verification of Configuration
- Submit test questions containing typical imaging report snippets, device logs, and maintenance records. Check if the returned citations include complete device information, key diagnostic terms, and timestamps.
- Perform batch upload tests with different formats of imaging reports and device log files. Confirm that all file types are successfully parsed and indexed.
- Regularly check the knowledge base index update status. Ensure consistency with the actual imaging report system's update frequency by comparing the ingestion time of the latest reports.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.