Face swapping in seizure videos for patient deidentification.

OBJECTIVE: This study aimed to test different AI-based face-swapping models applied to videos of epileptic seizures, with the goal of protecting patient privacy while retaining clinically useful seizure semiology. We hypothesized that specific models would show differences in semiologic fidelity compared to the original clinical videos. METHODS: Three open-source models, SimSwap, MobileFaceSwap and GHOST were adopted for face-swapping. For every model, an AI generated male and female image were used to replace the original faces. One representative seizure per patient from three patients with epilepsy was chosen (3 seizure videos x 3 AI models x 2 M/F swap) and remade to 18 transformed video clips. To evaluate the performance of the three models, we used both objective (AI-based) and subjective (expert clinician) evaluation. The objective assessment included four metrics for facial appearance and four metrics for facial expression changes. Four experienced epileptologists reviewed the clips and scoring according to deidentification and preservation of semiology. Kruskal-Wallis H test was used for statistical analysis among the models. RESULTS: In the reproduced videos, the swapped face cannot be recognized as the original face, with no significant difference in scores of deidentification either by objective or subjective assessment. Regarding semiology preservation, no significant differences between models were observed in the objective evaluations. The subjective evaluations revealed that the GHOST model outperformed the other two models (p=0.028). CONCLUSION: This is the first study evaluating AI face swapping models in epileptic seizure video clips. Optimization of AI face-swapping models could enhance the accessibility of seizure videos for education and research while protecting patient privacy and maintaining semiology.

Automated video analysis of emotion and dystonia in epileptic seizures.

OBJECTIVE: To investigate the accuracy of deep learning methods applied to seizure video data, in discriminating individual semiologic features of dystonia and emotion in epileptic seizures. METHODS: A dataset of epileptic seizure videos was used from patients explored with stereo-EEG for focal pharmacoresistant epilepsy. All patients had hyperkinetic (HKN) seizures according to ILAE definition. Presence or absence of (1) dystonia and (2) emotional features in each seizure was documented by an experienced clinician. A deep learning multi-stream model with appearance and skeletal keypoints, face and body information, using graph convolutional neural networks, was used to test discrimination of dystonia and emotion. Classification accuracy was assessed using a leave-one-subject-out analysis. RESULTS: We studied 38 HKN seizure videos in 19 patients. By visual analysis based on ILAE criteria, 9/19 patients were considered to have dystonia and 9/19 patients were considered to have emotional signs. Two patients had both dystonia and emotional signs. Applying the deep learning multistream model, spatiotemporal features of facial appearance showed best accuracy for emotion detection (F1 score 0.84), while skeletal keypoint detection performed best for dystonia (F1 score 0.83). SIGNIFICANCE: Here, we investigated deep learning of video data for analyzing individual semiologic features of dystonia and emotion in hyperkinetic seizures. Automated classification of individual semiologic features is possible and merits further study.

Neural correlates of rhythmic rocking in prefrontal seizures.

OBJECTIVES: Rhythmic body rocking movements may occur in prefrontal epileptic seizures. Here, we compare quantified time-evolving frequency of stereotyped rocking with signal analysis of intracerebral electroencephalographic data. METHODS: In a single patient, prefrontal seizures with rhythmic anteroposterior body rocking recorded on stereoelectroencephalography (SEEG) were analyzed using fast Fourier transform, time-frequency decomposition and phase amplitude coupling, with regards to quantified video data. Comparison was made with seizures without rocking in the same patient, as well as resting state data. RESULTS: Rocking movements in the delta (∼1 Hz) range began a few seconds after SEEG onset of low voltage fast discharge. During rocking movements: (1) presence of a peak of delta band activity was visible in bipolar montage, with maximal power in epileptogenic zone and corresponding to mean rocking frequency; (2) correlation, using phase amplitude coupling, was shown between the phase of this delta activity and high-gamma power in the epileptogenic zone and the anterior cingulate region. CONCLUSIONS: Here, delta range rhythmic body rocking was associated with cortical delta oscillatory activity and phase-coupled high-gamma energy. These results suggest a neural signature during expression of motor semiology incorporating both temporal features associated with rhythmic movements and spatial features of seizure discharge.

Rhythmic rocking stereotypies in frontal lobe seizures: A quantified video study.

OBJECTIVES: Rhythmic, stereotyped movements occur in some epileptic seizures. We aimed to document time-evolving frequencies of antero-posterior rocking occurring during prefrontal seizures, using a quantitative video analysis. METHODS: Six seizures from 3 patients with prefrontal epilepsy yet different sublobar localizations were analyzed using a deep learning-based head-tracking method. RESULTS: Mean rocking frequency varied between patients and seizures (0.37-1.0Hz). Coefficient of variation of frequency was low (≤12%). DISCUSSION: Regularity of body rocking movements suggests a mechanism involving intrinsic oscillatory generators. Since localization of seizure onset varied within prefrontal cortex across patients, altered dynamics converging on a "final common pathway" of seizure propagation involving cortico-subcortical circuits is hypothesized.