Developing MRI Epilepsy Lesion Detection (MELD) software for clinical use

Research questions Can we develop an AI-based MRI Epilepsy Lesion Detection tool (MELD) that is safe, effective and can be used in the clinical diagnosis of patients with epilepsy? Can MELD reduce inequalities in epilepsy diagnosis across the UK through supporting general radiologists to review scans at the level of expert neuroradiologists? Background Each year in the UK, 30,000 new epilepsy MRI scans are carried out. 20-30% of scans will have epilepsy-associated structural brain abnormalities. Scans are initially reported by general radiologists and 30-50% of epilepsy lesions are missed. Patients with missed lesions face years of poorly controlled seizures and additional investigations. In contrast, 70-80% of patients with MRI-diagnosed lesions who have epilepsy surgery are seizure free. The James Lind Alliance set “using artificial intelligence to aid diagnosis” as a top 10 priority. However, there are no regulated medical devices that can detect the range of structural lesions that cause focal epilepsy. MELD MELD provides AI-powered lesion detection. MELD is trained using labels created by expert radiologists, with the goal of enabling general radiologists to perform at the level of an expert. It aims to increase detection of epilepsy-causing lesions at first epilepsy MRI scan, reduce radiology reporting time and streamline pathways towards epilepsy surgery. Aims and objectives This project aims to develop MELD to have the clinical evidence (TRL7) and technical documentation required for MHRA submission as a Class IIb medical device, along with supporting health economic analysis for NICE evaluation. Methods The project is split into 6 work packages (WPs) to develop MELD for clinical use. The MELD prototype is a deep-learning algorithm based on a convolutional neural network. WP1: Preparation for clinical evaluation. Including: Optimisation of MELD software, packaging MELD into deployment-ready app, preparation for retrospective clinical evaluation and ethics application for clinical investigation of a medical device. WP2: Randomised clinical evaluation comparing diagnosis of epilepsy lesions with radiologists reporting first epilepsy MRI with and without access to MELD. WP3: Development of MELD into hospital-integrated end-to-end pipeline and pilot incorporation of MELD into clinical practice at GSTT. WP4: Develop Quality Management System and Medical Device File for MELD. WP5: Health Economic Analysis to quantify MELD’s value proposition and cost-effectiveness. WP6: Integration of patient perspectives in all objectives including co-development of MELD reports for accessible communication of software results. Timelines for delivery At project end we will have an optimised MELD software alongside the clinical evidence and technical documentation necessary for medical device regulation and commercialisation. Knowledge mobilisation, dissemination and impact Findings will be presented at scientific conferences, in peer-reviewed publications as well as to patient communities, via the patient expert group, Brain Buddy, EpiCARE SHAPE network and Young Epilepsy. We will also work with key partners within the NHS and medical technology sector to ensure efficient and effective deployment. Research inclusion The design and ongoing development of this project has and will involve diverse stakeholders, including clinicians and researchers (internationally via MELD network) and patients and parents (PPI expert group and Brain Buddy).