Balancing Clinical and Pathologic Relevence in the Machine Learning Diagnosis of Epilepsy

Nuclear imaging for localization and surgical outcome prediction in epilepsy: A review of latest discoveries and future perspectives

Background

Epilepsy is one of the most common neurological disorders. Approximately, one-third of patients with epilepsy have seizures refractory to antiepileptic drugs and further require surgical removal of the epileptogenic region. In the last decade, there have been many recent developments in radiopharmaceuticals, novel image analysis techniques, and new software for an epileptogenic zone (EZ) localization.

Objectives

Recently, we provided the latest discoveries, current challenges, and future perspectives in the field of positron emission tomography (PET) and single-photon emission computed tomography (SPECT) in epilepsy.

Methods

We searched for relevant articles published in MEDLINE and CENTRAL from July 2012 to July 2022. A systematic literature review based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis was conducted using the keywords "Epilepsy" and "PET or SPECT." We included both prospective and retrospective studies. Studies with preclinical subjects or not focusing on EZ localization or surgical outcome prediction using recently developed PET radiopharmaceuticals, novel image analysis techniques, and new software were excluded from the review. The remaining 162 articles were reviewed.

Results

We first present recent findings and developments in PET radiopharmaceuticals. Second, we present novel image analysis techniques and new software in the last decade for EZ localization. Finally, we summarize the overall findings and discuss future perspectives in the field of PET and SPECT in epilepsy.

Conclusion

Combining new radiopharmaceutical development, new indications, new techniques, and software improves EZ localization and provides a better understanding of epilepsy. These have proven not to only predict prognosis but also to improve the outcome of epilepsy surgery.

Clinical Application of Machine Learning Models for Brain Imaging in Epilepsy: A Review

Epilepsy is a common neurological disorder characterized by recurrent and disabling seizures. An increasing number of clinical and experimental applications of machine learning (ML) methods for epilepsy and other neurological and psychiatric disorders are available. ML methods have the potential to provide a reliable and optimal performance for clinical diagnoses, prediction, and personalized medicine by using mathematical algorithms and computational approaches. There are now several applications of ML for epilepsy, including neuroimaging analyses. For precise and reliable clinical applications in epilepsy and neuroimaging, the diverse ML methodologies should be examined and validated. We review the clinical applications of ML models for brain imaging in epilepsy obtained from a PubMed database search in February 2021. We first present an overview of typical neuroimaging modalities and ML models used in the epilepsy studies and then focus on the existing applications of ML models for brain imaging in epilepsy based on the following clinical aspects: (i) distinguishing individuals with epilepsy from healthy controls, (ii) lateralization of the temporal lobe epilepsy focus, (iii) the identification of epileptogenic foci, (iv) the prediction of clinical outcomes, and (v) brain-age prediction. We address the practical problems and challenges described in the literature and suggest some future research directions.

Machine learning for detecting mesial temporal lobe epilepsy by structural and functional neuroimaging

Mesial temporal lobe epilepsy (mTLE), the most common type of focal epilepsy, is associated with functional and structural brain alterations. Machine learning (ML) techniques have been successfully used in discriminating mTLE from healthy controls. However, either functional or structural neuroimaging data are mostly used separately as input, and the opportunity to combine both has not been exploited yet. We conducted a multimodal ML study based on functional and structural neuroimaging measures. We enrolled 37 patients with left mTLE, 37 patients with right mTLE, and 74 healthy controls and trained a support vector ML model to distinguish them by using each measure and the combinations of the measures. For each single measure, we obtained a mean accuracy of 74% and 69% for discriminating left mTLE and right mTLE from controls, respectively, and 64% when all patients were combined. We achieved an accuracy of 78% by integrating functional data and 79% by integrating structural data for left mTLE, and the highest accuracy of 84% was obtained when all functional and structural measures were combined. These findings suggest that combining multimodal measures within a single model is a promising direction for improving the classification of individual patients with mTLE.

Learning stable and predictive network-based patterns of schizophrenia and its clinical symptoms

Schizophrenia is often associated with disrupted brain connectivity. However, identifying specific neuroimaging-based patterns pathognomonic for schizophrenia and related symptom severity remains a challenging open problem requiring large-scale data-driven analyses emphasizing not only statistical significance but also stability across multiple datasets, contexts and cohorts. Accurate prediction on previously unseen subjects, or generalization, is also essential for any useful biomarker of schizophrenia. In order to build a predictive model based on functional network feature patterns, we studied whole-brain fMRI functional networks, both at the voxel level and lower-resolution supervoxel level. Targeting Auditory Oddball task data on the FBIRN fMRI dataset (n = 95), we considered node-degree and link-weight network features and evaluated stability and generalization accuracy of statistically significant feature sets in discriminating patients vs.

CONTROLS

We also applied sparse multivariate regression (elastic net) to whole-brain functional connectivity features, for the first time, to derive stable predictive features for symptom severity. Whole-brain link-weight features achieved 74% accuracy in identifying patients and were more stable than voxel-wise node-degrees. Link-weight features predicted severity of several negative and positive symptom scales, including inattentiveness and bizarre behavior. The most-significant, stable and discriminative functional connectivity changes involved increased correlations between thalamus and primary motor/primary sensory cortex, and between precuneus (BA7) and thalamus, putamen, and Brodmann areas BA9 and BA44. Precuneus, along with BA6 and primary sensory cortex, was also involved in predicting severity of several symptoms. Overall, the proposed multi-step methodology may help identify more reliable multivariate patterns allowing for accurate prediction of schizophrenia and its symptoms severity.

Multimodal diagnosis of epilepsy using conditional dependence and multiple imputation

The definitive diagnosis of the type of epilepsy, if it exists, in medication-resistant seizure disorder is based on the efficient combination of clinical information, long-term video-electroencephalography (EEG) and neuroimaging. Diagnoses are reached by a consensus panel that combines these diverse modalities using clinical wisdom and experience. Here we compare two methods of multimodal computer-aided diagnosis, vector concatenation (VC) and conditional dependence (CD), using clinical archive data from 645 patients with medication-resistant seizure disorder, confirmed by video-EEG. CD models the clinical decision process, whereas VC allows for statistical modeling of cross-modality interactions. Due to the nature of clinical data, not all information was available in all patients. To overcome this, we multiply-imputed the missing data. Using a C4.5 decision tree, single modality classifiers achieved 53.1%, 51.5% and 51.1% average accuracy for MRI, clinical information and FDG-PET, respectively, for the discrimination between non-epileptic seizures, temporal lobe epilepsy, other focal epilepsies and generalized-onset epilepsy (vs. chance, p<0.01). Using VC, the average accuracy was significantly lower (39.2%). In contrast, the CD classifier that classified with MRI then clinical information achieved an average accuracy of 58.7% (vs. VC, p<0.01). The decrease in accuracy of VC compared to the MRI classifier illustrates how the addition of more informative features does not improve performance monotonically. The superiority of conditional dependence over vector concatenation suggests that the structure imposed by conditional dependence improved our ability to model the underlying diagnostic trends in the multimodality data.