Classification and transfer learning of sleep spindles based on convolutional neural networks.

Background: Sleep plays a critical role in human physiological and psychological health, and electroencephalography (EEG), an effective sleep-monitoring method, is of great importance in revealing sleep characteristics and aiding the diagnosis of sleep disorders. Sleep spindles, which are a typical phenomenon in EEG, hold importance in sleep science. Methods: This paper proposes a novel convolutional neural network (CNN) model to classify sleep spindles. Transfer learning is employed to apply the model trained on the sleep spindles of healthy subjects to those of subjects with insomnia for classification. To analyze the effect of transfer learning, we discuss the classification results of both partially and fully transferred convolutional layers. Results: The classification accuracy for the healthy and insomnia subjects' spindles were 93.68% and 92.77%, respectively. During transfer learning, when transferring all convolutional layers, the classification accuracy for the insomnia subjects' spindles was 91.41% and transferring only the first four convolutional layers achieved a classification result of 92.80%. The experimental results demonstrate that the proposed CNN model can effectively classify sleep spindles. Furthermore, the features learned from the data of the normal subjects can be effectively applied to the data for subjects with insomnia, yielding desirable outcomes. Discussion: These outcomes underscore the efficacy of both the collected dataset and the proposed CNN model. The proposed model exhibits potential as a rapid and effective means to diagnose and treat sleep disorders, thereby improving the speed and quality of patient care.

EEG-FRM: a neural network based familiar and unfamiliar face EEG recognition method.

Recognizing familiar faces holds great value in various fields such as medicine, criminal investigation, and lie detection. In this paper, we designed a Complex Trial Protocol-based familiar and unfamiliar face recognition experiment that using self-face information, and collected EEG data from 147 subjects. A novel neural network-based method, the EEG-based Face Recognition Model (EEG-FRM), is proposed in this paper for cross-subject familiar/unfamiliar face recognition, which combines a multi-scale convolutional classification network with the maximum probability mechanism to realize individual face recognition. The multi-scale convolutional neural network extracts temporal information and spatial features from the EEG data, the attention module and supervised contrastive learning module are employed to promote the classification performance. Experimental results on the dataset reveal that familiar face stimuli could evoke significant P300 responses, mainly concentrated in the parietal lobe and nearby regions. Our proposed model achieved impressive results, with a balanced accuracy of 85.64%, a true positive rate of 73.23%, and a false positive rate of 1.96% on the collected dataset, outperforming other compared methods. The experimental results demonstrate the effectiveness and superiority of our proposed model.

Development of a maxillofacial virtual surgical system based on biomechanical parameters of facial soft tissue.

PURPOSE: Lack of biomechanical force model of soft tissue hinders the development of virtual surgical simulation in maxillofacial surgery. In this study, a physical model of facial soft tissue based on real biomechanical parameters was constructed, and a haptics-enabled virtual surgical system was developed to simulate incision-making process on facial soft tissue and to help maxillofacial surgery training. METHODS: CT data of a 25-year-old female patient were imported into Mimics software to reconstruct 3D models of maxillofacial soft and skeletal tissues. 3dMD stereo-photo of the patient was fused on facial surface to include texture information. Insertion and cutting parameters of facial soft tissue measured on fresh cadavers were integrated, and a maxillofacial biomechanical force model was established. Rapid deformation and force feedback were realized through localized deformation algorithm and axis aligned bounding box (AABB)-based collision detection. The virtual model was validated quantitatively and qualitatively. RESULTS: A patient-specific physical model composed of skeletal and facial soft tissue was constructed and embedded in the virtual surgical system. Insertion and cutting in different regions of facial soft tissue were simulated using omega 6, and real-time feedback force was recorded. The feedback force was consistent with acquired force data of experiments conducted on tissue specimen. Real-time graphic and haptic feedback were realized. The mean score of the system performance was 3.71 given by surgeons in evaluation questionnaires. CONCLUSION: The maxillofacial physical model enabled operators to simulate insertion and cutting on facial soft tissue with realization of realistic deformation and haptic feedback. The combination of localized deformation algorithm and AABB-based collision detection improved computational efficiency. The proposed virtual surgical system demonstrated excellent performance in simulation and training of incision-making process.

Accuracy assessment of a novel semiautomatic method evaluating bone grafts around the dental implant: an in vitro and ex vivo study.

The present study aimed to evaluate the accuracy and repeatability of morphological contour interpolation (MCI)-based semiautomatic segmentation method for volumetric measurements of bone grafts around dental implants. Three in vitro (one with a cylinder and two with a geometrically complex form) and four ex vivo models (peri-implant cylinder-shaped bone defect) were created for imitating implant placement with simultaneous guided bone regeneration (GBR) procedure. Cone beam computerized tomography (CBCT) scans of all models were obtained with the same parameters. For volumetric measurements, the actual volumes of bone grafts in models were assessed by computer-aided calculation and both manual and MCI-based methods were utilized as test methods. The accuracy of the methods was evaluated by comparing the measured value and the actual volume. The repeatability was assessed by calculating the coefficients of variation of repeated measurements. For the accuracy of three dimensional (3D) reconstructions, the computer-designed corresponding models were set as the reference and the morphological deviation of 3D surface renderings created by two methods were evaluated by comparing with reference. Besides, measurement time was recorded and a comparison between the two methods was performed. High accuracy of the MCI-based segmentation method was found with a discrepancy between the measured value and actual value never exceeding - 7.5%. The excellent repeatability was shown with coefficients of variation never exceeding 1.2%. The MCI-based method showed less measurement time than the manual method and its 3D surface rendering showed a lower deviation from the reference.