Mock review exercise 1
Project summary
This project deploys an artificial intelligence system across a consortium of European hospitals to support radiologists in chest X-ray interpretation and reporting. The system is designed to detect common radiological findings and, in particular, to identify high-confidence non-pathological images.
It is introduced with the aim of improving reporting speed, reducing routine workload, and increasing consistency in interpretation. The consortium also expects the system to reduce overlooked findings and allow radiologists to focus on complex cases.
Technical design
The AI system is used in three ways.
For chest X-rays classified as “high-confidence non-pathological”, the system automatically generates a draft report when model confidence is above 0.95. The radiologist may accept this report with one click. The consortium expects that up to 40% of reports may eventually be handled in this accelerated way. For these cases, the interface shows only the confidence score and the draft report, not highlighted image regions or interpreted findings.
For images classified as “abnormal”, the AI system functions as a diagnostic support tool. The interface displays highlighted suspicious regions, a confidence score, and a short list of likely findings to support review by the radiologist.
After the radiologist signs the final report, the system compares the human report with the AI-generated interpretation and flags substantial discrepancies for secondary review. However, hospital management has expressed concern that discrepancy-based review may create a growing review queue and increase turnaround times. At some hospitals, there is discussion of reducing the number of cases sent for secondary review in order to preserve efficiency gains.
Data handling
Training and validation data come from historical chest X-rays and associated radiology reports collected at participating hospitals. The consortium states that data processing will comply with the GDPR and the Medical Device Regulation.
The system logs outputs, confidence scores, overlays, and user interactions for quality assurance and audit purposes. Radiologists receive a short training session before deployment. In some hospitals, AI-generated reports may become visible to patients through the patient portal before a clinician has discussed the result with them.
Reviewer’s task
- Using the Checklist, identify serious and complex ethics issues and map them under the relevant ALTAI requirements.
- Using the Portfolio, propose up to 8 measures to address these issues.
- Detect potential ethics tension(s) in between the ALTAI ethics requirements.
Source: A. Prégent and A. Grinbaum, Developing Competency in AI Ethics: A training framework. arxiv:2610.13319