Pipeline evaluation of a state-of-the-art AI algorithm for detection of focal cortical dysplasia: insights into potential failure sources.

The purpose of this study is to investigate if multimodality neuroimaging evaluation increases the detection of subtle focal cortical dysplasia as part of an epilepsy surgery evaluation. Three patients with normal magnetic resonance imaging and histopathological findings of focal cortical dysplasia were reviewed. Their magnetoencephalography recordings were performed on whole-head magnetoencephalography system. Magnetic resonance images were re-evaluated with special inspection in limited regions guided by magnetoencephalography spike localization. Two patients had ictal and interictal single photon emission computed tomography study after administration of Tc99m ECD. In two patients we found tiny focal abnormalities including slightly increased cortical thickness and blurred gray–white matter junction at the locations of interictal events after re-evaluation of the MR images indicating focal cortical dysplasia. The third patient showed focal atrophic change. All patients are seizure free after surgery. Both ictal and interictal single photon emission computed tomography showed hyperperfusion in the dysplastic cortex regions. Multimodality neuroimaging study can improve the detection of focal cortical dysplasia. Normal magnetic resonance images should be re-evaluated for subtle signs of focal cortical dysplasia especially when magnetoencephalography recording demonstrate focal epileptic discharges.

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.

We aim to improve focal cortical dysplasia (FCD) detection by combining high-resolution, three-dimensional (3D) magnetic resonance fingerprinting (MRF) with voxel-based morphometric magnetic resonance imaging (MRI) analysis. We included 37 patients with pharmacoresistant focal epilepsy and FCD (10 IIa, 15 IIb, 10 mild Malformation of Cortical Development [mMCD], and 2 mMCD with oligodendroglial hyperplasia and epilepsy [MOGHE]). Fifty-nine healthy controls (HCs) were also included. 3D lesion labels were manually created. Whole-brain MRF scans were obtained with 1 mm3 isotropic resolution, from which quantitative T1 and T2 maps were reconstructed. Voxel-based MRI postprocessing, implemented with the morphometric analysis program (MAP18), was performed for FCD detection using clinical T1w images, outputting clusters with voxel-wise lesion probabilities. Average MRF T1 and T2 were calculated in each cluster from MAP18 output for gray matter (GM) and white matter (WM) separately. Normalized MRF T1 and T2 were calculated by z-scores using HCs. Clusters that overlapped with the lesion labels were considered true positives (TPs); clusters with no overlap were considered false positives (FPs). Two-sample t-testswere performed to compare MRF measures between TP/FP clusters. A neural network model was trained using MRF values and cluster volume to distinguish TP/FP clusters. Ten-fold cross-validation was used to evaluate model performance at the cluster level. Leave-one-patient-out cross-validation was used to evaluate performance at the patient level. MRF metrics were significantly higher in TP than FP clusters, including GM T1, normalized WM T1, and normalized WM T2. The neural network model with normalized MRF measures and cluster volume as input achieved mean area under the curve (AUC) of .83, sensitivity of 82.1%, and specificity of 71.7%. This model showed superior performance over direct thresholding of MAP18 FCD probability map at both the cluster and patient levels, eliminating ≥75% FP clusters in 30% of patients and ≥50% of FP clusters in 91% of patients. This pilot study suggests the efficacy of MRF for reducing FPs in FCD detection, due to its quantitative values reflecting invivo pathological changes. © 2024 International League Against Epilepsy.

The detection of focal cortical dysplasia (FCD) in magnetic resonance imaging is challenging. Voxel-based morphometric analysis and automated FCD detection using an artificial neural network (ANN) integrated into the Morphometric Analysis Program (MAP18) have been shown to facilitate FCD detection. This study aimed to evaluate whether the detection of FCD can be further improved by feeding this approach with magnetization prepared two rapid acquisition gradient echoes (MP2RAGE) instead of magnetization-prepared rapid acquisition gradient echo (MPRAGE) datasets. MPRAGE and MP2RAGE datasets were acquired in a consecutive sample of 32 patients with FCD and postprocessed using MAP18. Visual analysis and, if available, histopathology served as the gold standard for assessing the sensitivity and specificity of FCD detection. Out-of-sample specificity was evaluated in a cohort of 32healthy controls. The sensitivity and specificity of FCD detection were 82.4% and 62.5% for the MPRAGE and 97.1% and 34.4% for the MP2RAGE sequences, respectively. Median volumes of true-positive voxel clusters were .16ml for the MPRAGE and .52ml for the MP2RAGE sequences compared to .08- and .04-ml volumes of false-positive clusters. With regard to cluster volumes, FCD detection was substantially improved for the MP2RAGE data when the estimated optimal threshold of .23ml was applied (sensitivity = 72.9%, specificity = 83.0%). In contrast, the estimated optimal threshold of .37ml for the MPRAGE data did not improve FCD lesion detection (sensitivity = 42.9%, specificity = 79.5%). In this study, the sensitivity of FCD detection by morphometric analysis and an ANN integrated into MAP18 was higher for MP2RAGE than for MPRAGE sequences. Additional usage of cluster volume information helped to discriminate between true- and false-positive MP2RAGE results.

In this paper, a computer aided diagnosis (CAD) system for automatic detection of focal cortical dysplasia (FCD) on T1-weighted MRI is proposed. We introduce a new set of differential cluster-wise features comparing local differences of the candidate lesional area with its surroundings and other GM/WM boundaries. The local differences are measured in a distributional sense using χ² distances. Finally, a Support Vector Machine (SVM) classifier is used to classify the clusters. Experimental results show an 88% lesion detection rate with only 1.67 false positive clusters per subject. Also, the results show that using additional differential features clearly outperforms the result using only absolute features.

Focal cortical dysplasia (FCD), i.e., neuronal derangement due to developmental malformation, is increasingly recognized as an underlying cause of formerly cryptogenic focal epilepsy. However, in subtle cases, its diagnosis by visual evaluation of magnetic resonance images (MRI) remains difficult. Here, we present three novel techniques for postprocessing of 3-dimensional (3D) MRI which may improve lesion detection by enhancing image properties not readily accessible by visual analysis. Following the principles of voxel-based morphometry a T1-weighted MRI volume data set (MPRAGE) is normalized and segmented using algorithms of SPM99 (Statistical Parametric Mapping, Wellcome Department of Imaging Neuroscience, London). Then, the distribution of gray and white matter is analyzed on a voxel-wise basis and compared with a normal database consisting of the MR images of 53 healthy subjects. Based on this analysis, 3-dimensional maps called 'thickness image', 'extension image', and 'junction image', are created which characterize three different features of FCD, i.e., abnormal thickness of the cortical ribbon, abnormal extension of gray matter into the white matter, and blurring of the gray-white matter junction. These methods were applied to the MRI data of 25 epilepsy patients with histologically proven FCD. In each of the new feature maps the locations of the five highest maxima (corresponding to the maximum deviations from the mean of the normal database) were automatically determined and compared with the sites of the lesions in the conventional MR images or-in case of cryptogenic epilepsy-with the resection areas in the post-operative MRI. This approach was able to detect 15/25 lesions in the thickness image and 18/25 lesions in the junction and extension image, respectively. With all feature maps combined, 23 out of 25 dysplastic lesions were detected. Among these cases there were also four patients in whom the dysplastic lesion itself or at least an essential part of it had not been recognized on conventional MR images despite acquisition and assessment in a tertiary epilepsy center. The novel techniques for automated post-processing of MRI presented here facilitate the detection and localization of FCD and increase the sensitivity of MR imaging. Thereby, they provide a valuable additional diagnostic tool in the presurgical evaluation of epilepsy patients and improve the therapeutic options especially in cases of cryptogenic epilepsy.

SummaryObjectiveThe aim of this study is to determine whether the use of 7 tesla (T) MRI in clinical practice leads to higher detection rates of focal cortical dysplasias in possible candidates for epilepsy surgery.MethodsIn our center patients are referred for 7 T MRI if lesional focal epilepsy is suspected, but no abnormalities are detected at one or more previous, sufficient‐quality lower‐field MRI scans, acquired with a dedicated epilepsy protocol, or when concealed pathology is suspected in combination with MR‐visible mesiotemporal sclerosis—dual pathology. We assessed 40 epilepsy patients who underwent 7 T MRI for presurgical evaluation and whose scans (both 7 T and lower field) were discussed during multidisciplinary epilepsy surgery meetings that included a dedicated epilepsy neuroradiologist. We compared the conclusions of the multidisciplinary visual assessments of 7 T and lower‐field MRI scans.ResultsIn our series of 40 patients, multidisciplinary evaluation of 7 T MRI identified additional lesions not seen on lower‐field MRI in 9 patients (23%). These findings were guiding in surgical planning. So far, 6 patients underwent surgery, with histological confirmation of focal cortical dysplasia or mild malformation of cortical development.SignificanceSeven T MRI improves detection of subtle focal cortical dysplasia and mild malformations of cortical development in patients with intractable epilepsy and may therefore contribute to identification of surgical candidates and complete resection of the epileptogenic lesion, and thus to postoperative seizure freedom.

Focal cortical dysplasias are the most common resected epileptogenic lesions in children and the third most common lesion in adults, but they are often subtle and frequently overlooked on MR imaging. The purpose of this study was to evaluate whether MP2RAGE-based morphometric MR imaging analysis is superior to MPRAGE-based analysis in the detection of focal cortical dysplasia. MPRAGE and MP2RAGE datasets were acquired in a consecutive series of 640 patients with epilepsy. Datasets were postprocessed using the Morphometric Analysis Program to generate morphometric z score maps such as junction, extension, and thickness images based on both MPRAGE and MP2RAGE images. Focal cortical dysplasia lesions were manually segmented in the junction images, and volumes and mean z scores of the lesions were measured. Of 21 focal cortical dysplasias discovered, all were clearly visible on MP2RAGE junction images, whereas 2 were not visible on MPRAGE junction images. In all except 4 patients, the volume of the focal cortical dysplasia was larger and mean lesion z scores were higher on MP2RAGE junction images compared with the MPRAGE-based images (P = .005, P = .013). In this study, MP2RAGE-based morphometric analysis created clearer output maps with larger lesion volumes and higher z scores than the MPRAGE-based analysis. This new approach may improve the detection of subtle, otherwise overlooked focal cortical dysplasia.

PurposeTo evaluate a MRI postprocessing tool for the enhanced and rapid detection of focal cortical dysplasia (FCD).MethodsMP2RAGE sequences of 40 consecutive, so far MRI-negative patients and of 32 healthy controls were morphometrically analyzed to highlight typical FCD features. The resulting morphometric maps served as input for an artificial neural network generating a FCD probability map. The FCD probability map was inversely normalized, co-registered to the MPRAGE2 sequence, and re-transferred into the PACS system. Co-registered images were scrolled through “within a minute” to determine whether a FCD was present or not.ResultsFifteen FCD, three subcortical band heterotopias (SBH), and one periventricular nodular heterotopia were identified. Of those, four FCD and one SBH were only detected by MRI postprocessing while one FCD and one focal polymicrogryia were missed, respectively. False-positive results occurred in 21 patients and 22 healthy controls. However, true positive cluster volumes were significantly larger than volumes of false-positive clusters (p < 0.001). The area under the curve of the receiver operating curve was 0.851 with a cut-off volume of 0.05 ml best indicating a FCD.ConclusionAutomated MRI postprocessing and presentation of co-registered output maps in the PACS allowed for rapid (i.e., “within a minute”) identification of FCDs in our clinical setting. The presence of false-positive findings currently requires a careful comparison of postprocessing results with conventional MR images but may be reduced in the future using a neural network better adapted to MP2RAGE images.

ObjectiveThis study aims to report human performance in the detection of Focal Cortical Dysplasias (FCDs) using an openly available dataset. Additionally, it defines a subset of this data as a “difficult” test set to establish a public baseline benchmark against which new methods for automated FCD detection can be evaluated.MethodsThe performance of 28 human readers with varying levels of expertise in detecting FCDs was originally analyzed using 146 subjects (not all of which are openly available), we analyzed the openly available subset of 85 cases. Performance was measured based on the overlap between predicted regions of interest (ROIs) and ground‐truth lesion masks, using the Dice‐Soerensen coefficient (DSC). The benchmark test set was chosen to consist of 15 subjects most predictive for human performance and 13 subjects identified by at most 3 of the 28 readers.ResultsExpert readers achieved an average detection rate of 68%, compared to 45% for non‐experts and 27% for laypersons. Neuroradiologists detected the highest percentage of lesions (64%), while psychiatrists detected the least (34%). Neurosurgeons had the highest ROI sensitivity (0.70), and psychiatrists had the highest ROI precision (0.78). The benchmark test set revealed an expert detection rate of 49%.SignificanceReporting human performance in FCD detection provides a critical baseline for assessing the effectiveness of automated detection methods in a clinically relevant context. The defined benchmark test set serves as a useful indicator for evaluating advancements in computer‐aided FCD detection approaches.Plain Language SummaryFocal cortical dysplasias (FCDs) are malformations of cortical development and one of the most common causes of drug‐resistant focal epilepsy. Once found, FCDs can be neurosurgically resected, which leads to seizure freedom in many cases. However, FCDs are difficult to detect in the visual assessment of magnetic resonance imaging. A myriad of algorithms for automated FCD detection have been developed, but their true clinical value remains unclear since there is no benchmark dataset for evaluation and comparison to human performance. Here, we use human FCD detection performance to define a benchmark dataset with which new methods for automated detection can be evaluated.

Automated detection of focal cortical dysplasia based on magnetic resonance imaging and positron emission tomography

The purpose of this study is to perform an independent assessment of three state-of-the-art tools for the detection of focal cortical dysplasia (FCD) from Magnetic Resonance images (MRI). These tools include DeepFCD, the Multi-center Epilepsy Lesion Detection (MELD) Classifier, and MELDGraph. T1-weighted and fluid-attenuated inversion recovery MRIs from 101 epilepsy patients with FCD and 101 epilepsy patients without FCD were retrospectively included. Classifiers were evaluated at patient-level by their ability to correctly identify the presence of any FCD lesions, and at lesion-level by their capacity to identify lesions within regions delineated by neuroradiologists in MRI reports. A calibrated threshold for DeepFCD prediction probabilities was empirically determined to improve classifier specificity. Classifier test-retest consistency was measured using the Dice coefficient on repeated MRI scans of 21 individuals. At patient-level, MELDClassifier achieved 52% accuracy (sensitivity=91%, specificity=14%), MELDGraph reached 61% accuracy (sensitivity=76%, specificity=47%) and DeepFCD performed with 56% accuracy (sensitivity=62%, specificity=50%) at an empirically determined threshold of 0.90. At lesion-level, MELDClassifier performed with a sensitivity of 70% and a positive predictive value (PPV) of 13%. MELDGraph reached 53% sensitivity and PPV of 36%, whereas the DeepFCD performed with 30% sensitivity and PPV of 19%. Test-retest reliability was low, with an average [min, max] Dice coefficient of 0.28 [0.0, 1.0] for MELDClassifier, 0.38 [0.0, 1.0] for MELDGraph, and 0.35 [0.05, 0.54] for DeepFCD. This study highlights the current limitations of using deep learning models in FCD diagnosis and emphasizes the need to enhance the tools' accuracy, reliability, and interpretability to improve clinical utility.

Focal cortical dysplasia is a congenital abnormality of cortical development and the leading cause of surgically remediable drug-resistant epilepsy in children. Post-surgical outcome is improved by presurgical lesion detection on structural MRI. Automated computational techniques have improved detection of focal cortical dysplasias in adults but have not yet been effective when applied to developing brains. There is therefore a need to develop reliable and sensitive methods to address the particular challenges of a paediatric cohort.We developed a classifier using surface-based features to identify focal abnormalities of cortical development in a paediatric cohort. In addition to established measures, such as cortical thickness, grey-white matter blurring, FLAIR signal intensity, sulcal depth and curvature, our novel features included complementary metrics of surface morphology such as local cortical deformation as well as post-processing methods such as the “doughnut” method - which quantifies local variability in cortical morphometry/MRI signal intensity, and per-vertex interhemispheric asymmetry. A neural network classifier was trained using data from 22 patients with focal epilepsy (mean age = 12.1 ± 3.9, 9 females), after intra- and inter-subject normalisation using a population of 28 healthy controls (mean age = 14.6 ± 3.1, 11 females). Leave-one-out cross-validation was used to quantify classifier sensitivity using established features and the combination of established and novel features.Focal cortical dysplasias in our paediatric cohort were correctly identified with a higher sensitivity (73%) when novel features, based on our approach for detecting local cortical changes, were included, when compared to the sensitivity using only established features (59%). These methods may be applicable to aiding identification of subtle lesions in medication-resistant paediatric epilepsy as well as to the structural analysis of both healthy and abnormal cortical development.

Focal cortical dysplasia is one of the most common underlying pathologies in patients who undergo surgery for refractory epilepsy. Absence of a MRI-visible lesion necessitates additional diagnostic tests and is a predictor of poor surgical outcome. We describe a series of six patients with refractory epilepsy due to histopathologically-confirmed focal cortical dysplasia, for whom pre-surgical 7 tesla T2*-weighted MRI was acquired. In four of six patients, T2* sequences showed areas of marked superficial hypointensity, co-localizing with the epileptogenic lesion. 7 tesla T2* hypointensities overlying focal cortical dysplasia may represent leptomeningeal venous vascular abnormalities associated with the underlying dysplastic cortex. Adding T2* sequences to the MRI protocol may aid in the detection of focal cortical dysplasias.

Double Inversion Recovery Magnetic Resonance Imaging in Identifying Focal Cortical Dysplasia