Biomedical Engineering: Applications, Basis and Communications

Purpose: This study aimed to achieve two primary goals. First, we sought to develop a lightweight convolutional neural network (CNN) model, COLRECTSEG-UNet, for the purpose of automated polyp segmentation and localization. Second, we evaluated the generalizability of the proposed model by testing it on a proprietary colonoscopy image dataset (own dataset) in addition to a publicly available benchmark (CVC-ColonDB and Kvasir dataset). Methods: The COLRECTSEG-UNet architecture was employed to perform polyp segmentation. Subsequently, the model was trained and validated using an 80:20 data split from the CVC-ColonDB dataset. The efficacy of these data split was confirmed through $K$ -fold cross-validation. To elucidate the significance of each layer within the COLRECTSEG-UNet design, an ablation study was conducted. The performance of the model was assessed using various metrics on the CVC-ClinicDB, own dataset and Kvasir dataset during the testing phase. These metrics were then compared to those achieved by current state-of-the-art models and findings reported in recent literature. Results: When evaluated on the CVC-ClinicDB dataset, the COLRECTSEG-UNet model achieved outstanding accuracy of 0.9591, intersection over union (IoU) of 0.9299 and F1-Score of 0.9530. Similarly, impressive results were obtained on the own dataset, with accuracy, IoU and F1-Score values of 0.955, 0.9346 and 0.9539, respectively. Additionally, Kvasir dataset attained the accuracy, IoU and F1-Score of 0.9542, 0.9291 and 0.9565, respectively. These results demonstrate that the proposed COLRECTSEG-UNet outperforms the existing benchmark models and surpasses the performance reported in current literature. Conclusion: The implementation of COLRECTSEG-UNet as a lightweight model paves the way for its integration as a backend component within medical imaging system software. This has the potential to significantly aid gastroenterologists in clinical decision-making during interventions.

Brain activity leads to devastating effects on life which may lead to the loss of human lives. It can be detected at early stages to save human life. An electroencephalogram (EEG) is a test that will detect abnormalities in the brain wave. Electrodes are applied to the scalp during an EEG. These are tiny metal disks connected by slender wires. They pick up microscopic electrical charges produced by the brain’s cell activity. The results of an EEG reveal alterations in brain activity that may be helpful in the diagnosis of various brain disorders, particularly epilepsy and other conditions that result in seizures. In this research, a novel approach termed Tasmanian Devil Hunting Optimization-Deep Maxout Network (TDHO-DMN) is devised for brain activity detection based on motor imagery EEG signals. Initially, the input EEG signal obtained from the dataset is subjected to the signal pre-processing phase. Here, the input signals are pre-processed for denoising utilizing the Gaussian Filter. After that, the pre- processed signal is allowed for the feature extraction to extract the suitable feature vectors like amplitude modulation spectrum (AMS), frequency-based features and statistical features. Then, extracted features are fed to data augmentation which is carried out utilizing the oversampling technique. Finally, brain activity detection is accomplished by the Deep Maxout Network (DMN), which is trained by the Tasmanian Devil Hunting Optimization (TDHO) algorithm. TDHO is formed by the combination of Tasmanian Devil Optimization (TDO) and Deer Hunting Optimization Algorithm (DHOA). The performance evaluation of the proposed TDHO_DMN is analyzed using two benchmark datasets, where the proposed TDHO_DMN approach obtained a better performance in terms of accuracy, sensitivity and specificity of 90.70%, 91.00% and 91.40%, respectively.

The purpose of this study is to evaluate the sway areas of the center of pressure (COP) and the inclination angles between the center of mass (COM) and COP among elderly subjects with and without training in Chinese traditional exercises, including Tai Chi (TC) and Yuanji-Dance (YD). Additionally, this study investigates the electromyographic (EMG) activities during walking across these groups. The current study employed a cross-sectional study design. Twenty-seven healthy participants aged 45–70 years old were recruited from the local community in this study. TC group and YD group should have regular exercise training for at least two years, respectively. In the first part of this study, each subject of three groups (TC, YD and CON) was asked to walk at a self-selected speed on an 8 m walkway for gait analysis. Maximum center of mass–center of pressure (COM–COP) inclination angles in the sagittal plane and frontal plane were calculated. The EMG activity was recorded bilaterally from four muscles of the lower limbs including quadriceps, hamstring, tibialis anterior and gastrocnemius. In the second part, each subject was guided to perform four tandem static balance tests with the dominant leg leading on a force plate. The time-varying COP positions and the sway area of the COP were calculated from the ground reaction force data. A one-way ANOVA was conducted to compare walking variables between groups, and a two-way repeated measures ANOVA was used to evaluate the effects of group (between-group) and conditions (within-group) on the variables of the static tests. There were no significant differences in the maximum COM–COP inclination angles in anterior– posterior ($p$ = 0.06) and medial–lateral ($p$ = 0.15) directions as well as the EMG activities ($p$ > 0.05) among the TC, YD and CON groups. For static balance test, the TC group showed decreased excursion of COP in anterior– posterior direction and sway area compared with the CON group ($p$ < 0.01). For the YD group, decreased excursion of COP in the anterior–posterior direction during tandem stance on the soft surface was found ($p$ < 0.01). Moreover, the YD group showed decreased sway areas of COP compared with the CON group ($p$ < 0.01). The findings show that the TC group demonstrated smaller sway areas compared to the YD group, indicating that TC training may be especially effective in enhancing the static standing balance. However, there were no significant differences in the anterior–posterior and medial–lateral COM–COP inclination angles during walking among the TC, YD, and control groups. This result suggests that the training effects of TC and YD may not significantly impact the dynamic balance or the muscle activities observed during level walking.

COVID-19 is a respiratory disease affecting humans and animals. The disease has rapidly spread worldwide and became a pandemic in 2020. Preventing the virus from spreading has become increasingly challenging, especially with the need to test potential suspects rapidly. Deep learning-based methods have been developed to address this challenge of detecting COVID-19 from chest Computed Tomography (CT) images. The proposed network has multi-scale feature extraction layers with wavelet pooling. Learning features at different scales will enable the architecture to explore local patterns at different dimensions. So, in the proposed architecture, we have included a multi-scale convolutional layer to focus on sparse local regions about the disease conditions. Texture-based feature learning using wavelet pooling is incorporated into the architecture to improve detection performance. The proposed network achieved an accuracy of 99.79% with an AUC value of 0.9999. Compared with the existing methods, the proposed network has a lower computational cost regarding learnable parameters, FLOPS, and memory requirements. The proposed CNN model benefits from multi-scale structure and wavelet-pooling, resulting in superior performance compared to previous algorithms.

The acceptability of an electronic medical record (EMR) system as determined by the individual who uses it is critical to its effective adoption. EMRs increase public care as well as safety by shortening treatment times and eliminating medical mistakes. The primary goal is to examine how healthcare quality and utilization are affected by electronic health records (EHRs). This work presents evidence of EMR implementation and uptake by healthcare service providers. Survey responses from professionals and medical officials from the selected hospitals in Delhi were gathered and examined employing exploratory factor analysis and modeling of structural equations approaches to test our research model. Results of the study showed that system compatibility, perceived security, information quality, and adequate IT infrastructure favorably impact the quality and use of healthcare, but perceived usefulness unfavorably impacts the quality and use of healthcare. As a result of the research, the hospital ought to promote and preserve the EHR system.

Schizophrenia (SCH) is a psychiatric disorder that causes mental illness and is very difficult to discriminate from the normal subject. As the current diagnostic procedure DSM-V is qualitative, accurate clinical detection is essential. Cognitive dysfunction is widespread in $S$CH and it relates to brain activity. The aim of this research is to analyze EEG as a screening tool to identify SCH as it records brain activity. To stimulate cognitive activity, two different types of visual oddball paradigms are designed and are named P1 and P2. Although many researchers have analyzed the EEG of SCH, it remains unclear which recording condition is suitable. Therefore, EEG is recorded from 52 SCH subjects and 29 normal subjects during conventional recording conditions, namely rest, HV, PHV, Photic and cognitive activities (P1 and P2). The Welch power spectrum method is used to find the absolute power of the four EEG bands called $\delta$_Pr, $𝜃$_Pr, $\alpha$_Pr and $\beta$_Pr. In addition to these, each band’s peak power $\delta$_PPR, $𝜃$_PPR, $\alpha$_PPR and $\beta$_PPR and peak frequencies $\delta$_PF, $𝜃$_PF, $\alpha$_PF and $\beta$_PF are also calculated from the power spectrum of the EEG signal. Out of these 12 features, $\alpha$_Pr and $𝜃$_PF are considered as the most significant features, because they are significant with $p$ < 0.001 in all recording conditions. The SVM classifier successfully predicts SCH with a maximum sensitivity of 98% and accuracy of 97% using $\alpha$_Pr and $𝜃$_PF. Therefore, it can be concluded that $\alpha$_Pr and $𝜃$_PF can be used as features to use EEG as a screening tool for the identification of schizophrenia.

Optic disc and optic cup segmentation are needed to detect glaucoma and diabetic retinopathy disorders in the retina. The separation of optic discs, optic cups, and other features in retinal images is semantic segmentation. Semantic segmentation requires a robust and valid architecture to be able to separate features in images accurately. This study proposes the RIB-Net architecture. The RIB-Net architecture is a modified U-Net architecture with residual blocks in the bridge section and inception block in each layer in the encoder and decoder sections. Residual blocks serve to overcome the problem of vanishing gradients and enable more effective information flow in bridge paths in U-Net. Inception blocks help recognize features of varying sizes in the encoder and decoder sections. This modification provides an architecture that has good capabilities in the semantic segmentation of more complex image features. The performance results of the RIB-Net architecture are accuracy of 99.21%, IoU of 55.32%, F1-score of 70.95%, sensitivity of 63.84%, and specificity of 99.71%. Other performance results from the RIB-Net architecture are $\alpha$ kappa coefficient of 0.758, MCC of 0.764, and G-means of 0.82. The RIB-Net architecture has a good balance between sensitivity and specificity, so it can be concluded that RIB-Net works well for optic disc and optic cup segmentation in retinal images. The results of the F1-score, IOU, and sensitivity still need to be improved so RIB-Net can be developed as a model for developing an automatic retinal image segmentation system.

In response to the issue of insufficient information and low classification accuracy in single-channel feature extraction algorithms for EEG signals, the paper proposed a method based on Adaptive Noise-assisted Complete Ensemble Empirical Mode Decomposition (CEEMDAN) and multi-feature fusion. Firstly, we acquired multi-channel motor imagery EEG data and preprocessed it. Next, utilized CEEMDAN with adaptive noise to adaptively decompose the channel data. Secondly, extracted features from the Intrinsic Mode Functions (IMFs) which we obtained after CEEMDAN decomposition for different feature layers and fused them, obtaining the original feature matrices for each feature layer. Finally, we fused and reduced the dimensionality of the original feature matrices from each layer, input the resulting matrix into a classification model for classification, and output the classification results. We achieve an average classification accuracy of 88.59% on the BCI Competition II Data Set III, and validate the results on self-collected single-channel motor imagery EEG data, achieving a classification accuracy of 89.72%. Experimental results indicate that multi-feature fusion outperforms single features, and the combination of CEEMDAN decomposition with multi-feature fusion achieves higher classification accuracy.