Abstract:Background: The most common kind of focal epilepsy and the type that responds the least to pharmacological therapy is temporal lobe epilepsy (TLE). There is a high rate of misdiagnosis, and as a result, many patients are not treated correctly. The 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography/magnetic resonance imaging (PET/MRI) has recently been used for the evaluation of epilepsy. Despite this, the visual assessment makes it very difficult to identify accurately the onset of hypometabolic abnormalities. In this study, we developed an original, broad learning system based on symmetry for the identification of both the lobe and side epileptic foci in patients with TLE.Methods: We retrospectively enrolled 220 TLE patients (94 on the right and 126 on the left) with drug-resistant temporal lobe epilepsy, verified by pathological results, and 24 age-matched controls who underwent 18F-FDG PET and MRI scans. 18F-FDG PET/MRI imaging was quantitatively investigated using asymmetry features, and a broad learning system-based Siamese net (BLS-Siamese net) was proposed for precise discrimination of TLE patients as a binomial task, and then precise localization of epileptic foci was automatically calculated by a similarity algorithm as a multinomial task. In evaluating the effectiveness of the proposed framework, machine learning and visual evaluation methods were compared.Findings: The proposed BLS-Siamese net based on 18F-FDG PET/MRI imaging both exhibited excellent diagnostic performance for identifying TLE, with the area under the curve (AUC) of 0·966 and 0·969, the accuracy of 0·943 and 0·961, sensitivity of 0·979 and 0·988, specificity of 0·873 and 0·907, precision of 0·938 and 0·955 and F1 score of 0·958 and 0·971, in PET and T1WI imaging respectively. For the triple classification task in 18F-FDG PET and MRI imaging, the left and right-side detection accuracy and sensitivity are both higher than the other methods. The detection AUC of the BLS-Siamese net was 0·925 (left side) and 0·926 (right side) in PET imaging, and 0·783 (left side) and 0·748 (right side) in T1WI imaging. The proposed method performed significantly better in binomial and multinomial classification than visual assessment unblinded to clinical information.Interpretation: If validated in prospective cohorts, the original BLS-Siamese net based on 18F-FDG PET/MRI imaging could be used to easily and correctly detect epileptic foci, both in the lobe and side precisely, which perhaps utilized in decision-making for diagnosing epileptics in real clinical practice.Funding: The Clinical Frontier Technology Program of the First Affiliated Hospital of JinanUniversity, China (No. JNU1AF-CFTP-2022-a01214), the Project of Innovative Teamfor the Guangdong Universities (No. 2018KCXTD001), the National Natural ScienceFoundation of China (No. 81701751, No. 81871383), Guangdong Basic and AppliedBasic Research Foundation (No. 2020A1515011192, No. 2021A1515011999), theNational Key Research and Development Program of China (No. 2018YFC2002500),and Guangdong Provincial Key Laboratory of Traditional Chinese MedicineInformatization (No. 2021B1212040007).Declaration of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.Ethical Approval: We acquired the preoperative PET/MRI scans from the PACS (Picture Archiving and Communication System) at the PET CT/MRI center of the First Affiliated Hospital of Jinan University and the demographic data from Guangdong 999 Brain Hospital. This retrospective study was approved by the institution’s ethical committee, and informed consent was waived due to the retrospective nature of the study.

Enhancing the Diagnostic Utility of ASL Imaging in Temporal Lobe Epilepsy through FlowGAN: An ASL to PET Image Translation Framework

A RESOURCE-EFFICIENT DEEP LEARNING FRAMEWORK FOR LOW-DOSE BRAIN PET IMAGE RECONSTRUCTION AND ANALYSIS

A deep learning framework for 18 F-FDG PET imaging diagnosis in pediatric patients with temporal lobe epilepsy

A Novel Framework to Synthesize Arterial Spin Labeling Images Using Difference Images

Arterial Spin Labeling Images Synthesis from Smri Using Unbalanced Deep Discriminant Learning

A deep learning framework for 18F-FDG PET imaging diagnosis in pediatric patients with temporal lobe epilepsy

A new VAE-GAN model to synthesize arterial spin labeling images from structural MRI

A novel locally-constrained GAN-based ensemble to synthesize arterial spin labeling images

Arterial Spin Labeling Images Synthesis via Locally-Constrained WGAN-GP Ensemble

Exploring ASL perfusion MRI as a substitutive modality for 18F-FDG PET in determining the laterality of mesial temporal lobe epilepsy

Glucose Metabolic Profile by Visual Assessment Combined with Statistical Parametric Mapping Analysis in Pediatric Patients with Epilepsy.

Image synthesis of interictal SPECT from MRI and PET using machine learning

MRI to PET Cross-Modality Translation using Globally and Locally Aware GAN (GLA-GAN) for Multi-Modal Diagnosis of Alzheimer's Disease

Arterial spin labeling as a promising alternative to FDG-PET for clinical diagnosis of patients with disorders of consciousness

Cross-Modality Generation of Amyloid PET from FDG PET for Alzheimer’s Disease Diagnosis

A Broad Learning System for &Lt;sup>18</sup>f-Fdg PET/MRI Imaging Diagnosis in Temporal Lobe Epilepsy Patients: A Retrospective Cohort Study

Unsupervised arterial spin labeling image superresolution via multiscale generative adversarial network

ASL to PET Translation by a Semi-supervised Residual-based Attention-guided Convolutional Neural Network

A deep learning model for generating [18F]FDG PET Images from early-phase [18F]Florbetapir and [18F]Flutemetamol PET images

Cross-Modality Image Translation From Brain 18 F-FDG PET/CT Images to Fluid-Attenuated Inversion Recovery Images Using the CypixGAN Framework