Enhanced Readout of Imaging Communications with AI (ERICA)

  • Research type

    Research Study

  • Full title

    Enhanced Readout of Imaging Communications with AI (ERICA): An exploratory qualitative study of patient perspectives on the utility of GPT4-generated plain language summaries of glioblastoma MRI report texts

  • IRAS ID

    353054

  • Contact name

    Rajesh Jena

  • Contact email

    rjena@nhs.net

  • Sponsor organisation

    Cambridge University Hospitals NHS Foundation Trust

  • Duration of Study in the UK

    0 years, 6 months, 1 days

  • Research summary

    This study aims to assess how useful brain tumour patients find plain language summaries (PLS) of their brain Magnetic Resonance Imaging (MRI) reports, generated using the online large language model (LLM) ChatGPT4.

    MRI reports are pieces of writing created by radiologists. They describe MRI images using medical vocabulary and are designed for other clinicians to read and understand. Regular MRI scans are crucial to standard-of-care treatment and monitoring of glioblastoma, the commonest adult primary brain tumour. Within two weeks of each MRI, brain tumour patients at Cambridge University Hospitals (CUH) are offered an appointment to discuss key scan findings – such as whether their disease is stable or progressing - with a clinician at the neuro-oncology clinic. Afterwards, patients receive the complete, more complex MRI report electronically via MyChart. Reading this can cause additional concern and misunderstanding.

    LLMs including ChatGPT4 are a type of artificial intelligence (AI) that can generate PLS of specialist documents like MRI reports. PLS are summaries of complex texts, written using vocabulary the average person can understand. Our study makes a novel contribution to research into how this technology can be used to supplement patient encounters with clinicians, improving care by addressing unmet needs.

    Before the study, we tested ChatGPT4’s ability to generate useful summaries using synthetic MRI reports. We now plan to recruit forty glioblastoma patients from CUH’s neuro-oncology clinic, and seek their consent to upload a de-identified version of the latest brain MRI report they have discussed with a clinician. A non-radiologist clinician will assess the ChatGPT4-generated PLS for accuracy. Then, we will send it to patients via MyChart alongside an online questionnaire. This will gather patient perspectives on whether routinely adding a ChatGPT4-generated PLS could improve understanding of their MRI reports. We expect the study to last 24 weeks from recruitment to completion.

  • REC name

    East of England - Essex Research Ethics Committee

  • REC reference

    26/EE/0058

  • Date of REC Opinion

    31 Mar 2026

  • REC opinion

    Further Information Favourable Opinion