Interpretable surface-based detection of focal cortical dysplasias: a Multi-centre Epilepsy Lesion Detection study

Abstract

One outstanding challenge for machine learning in diagnostic biomedical imaging is algorithm interpretability. A key application is the identification of subtle epileptogenic focal cortical dysplasias (FCDs) from structural MRI. FCDs are difficult to visualize on structural MRI but are often amenable to surgical resection. We aimed to develop an open-source, interpretable, surface-based machine-learning algorithm to automatically identify FCDs on heterogeneous structural MRI data from epilepsy surgery centres worldwide. The Multi-centre Epilepsy Lesion Detection (MELD) Project collated and harmonized a retrospective MRI cohort of 1015 participants, 618 patients with focal FCD-related epilepsy and 397 controls, from 22 epilepsy centres worldwide. We created a neural network for FCD detection based on 33 surface-based features. The network was trained and cross-validated on 50% of the total cohort and tested on the remaining 50% as well as on 2 independent test sites. Multidimensional feature analysis and integrated gradient saliencies were used to interrogate network performance. Our pipeline outputs individual patient reports, which identify the location of predicted lesions, alongside their imaging features and relative saliency to the classifier. On a restricted ‘gold-standard’ subcohort of seizure-free patients with FCD type IIB who had T1 and fluid-attenuated inversion recovery MRI data, the MELD FCD surface-based algorithm had a sensitivity of 85%. Across the entire withheld test cohort the sensitivity was 59% and specificity was 54%. After including a border zone around lesions, to account for uncertainty around the borders of manually delineated lesion masks, the sensitivity was 67%. This multicentre, multinational study with open access protocols and code has developed a robust and interpretable machine-learning algorithm for automated detection of focal cortical dysplasias, giving physicians greater confidence in the identification of subtle MRI lesions in individuals with epilepsy.

Predicting seizure outcome after epilepsy surgery: do we need more complex models, larger samples, or better data?

“…Critically, data must be collected and curated in a standardized manner, as highlighted by experts 47 and similar to recent multicenter endeavours. 22,48,49 Here, it will be important to distinguish between investigating variables that may be predictive of outcome and identifying variables that can (feasibly) be included as predictors in a clinical decision-making tool. For the purpose of developing a clinical decision-making tool, we suggest including only variables that are routinely collected for all epilepsy surgery patients at most centers, to avoid introducing bias into the model.…”

Focal cortical dysplasia: a practical guide for neurologists

“…Seizure freedom after surgery is more likely if the black line region is included in resection 42 43. Machine-learning algorithms for automated detection of subtle FCD on neuroimaging are being developed 44…”

“…(42,43) Machine-learning algorithm for automated detection of subtle FCD on neuroimaging are being developed. (44) Additional functional imaging modalities, such as interictal fluorodeoxyglucose-positron emission tomography (FDG-PET) and subtraction of ictal/interictal single-photon emission computed tomography (SPECT) and its co-registration with structural MRI, may add important information in patients with subtle lesions that helps to increase the confidence of the structural MRI diagnosis. (45) • What is the histopathological presentation?…”

Towards Responsible Artificial Intelligence in Medicine: Reflections from the Australian Epilepsy Project

“…Contemporary AI solutions such as deep learning have been explicitly developed for image-based applications and are well suited for detecting neuroanatomical abnormalities 37,38 . An example of a transparent and explainable lesion detection framework is proposed by Spitzer et al 39 , namely the Multicentre Epilepsy Lesion Detection (MELD) framework. The MELD framework relies on detailed structural information about the brain to obtain a probability of a seizure focus based on structural brain imaging modalities commonly acquired in hospitals and research institutions.…”

“…In addition to lesion probability, MELD provides interpretable outputs regarding the brain's cortical thickness, curvature, and white and grey matter imaging contrast explaining the model's output. Spitzer et al 39 validated this approach with 1015 participants (epilepsy and controls) using a split-half training and testing paradigm. Patients with a visible brain lesion -focal cortical dysplasia-were detected with 85% accuracy.…”