Right Hemisphere Deep Exploration

[Skyy Mansour]

Thesite is secure.

The ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Background: This study aimed to investigate the association of superficial cerebral veins (SCVs) with sex-related cognitive differences and the possible hemodynamic mechanisms underlying these associations.

Methods: This investigation was a prospective case-control study. A total of 344 healthy volunteers were recruited. In all, 200 volunteers were included to establish the deep learning model, and 144 volunteers were used for the research, including 72 males (50%) and 72 females (50%). No significant differences in age (P=0.358) or education (P=0.779) were observed between the sexes. Cognitive functioning was evaluated using neuropsychological tests, including the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment-Basic (MOCA-B). Susceptibility-weighted imaging scans were acquired with a 3.0 T magnetic resonance imaging system using a 32-channel high-resolution phased array coil. Minimum intensity projection images were obtained by reconstructing susceptibility-weighted imaging data. A deep learning model was trained on the minimum intensity projection images to quantify the diameter, tortuosity index, length, and the number of SCVs in the bilateral cerebral hemispheres. Finally, the association between cognitive differences between males and females and the properties of the SCVs was analyzed.

Conclusions: The cognitive function of males was better than that of females, and the different numbers of SCVs may be one of the explanations for this phenomenon of sex-based differences in cognition.

Both the prevalence of certain diseases and their manifestations and treatment efficacy may differ depending on the sex of the patient. Thus, the identification of populations with particular individual characteristics is important to improve clinical prevention and treatment effects (1-3). It is well-established that the performance of cognition is related to many factors, which can be divided into social and biological factors. As many studies have pointed to the differences in cognition between sexes, the relevant pathophysiological mechanisms underlying sex-related differences in thinking and cognitive performance have generated considerable research interest and attention, particularly the differences in brain anatomy between the sexes. Sex differences of the brain have been observed in both humans and mice. For example, the brain volumes of healthy males are significantly larger than those of females (4-7). Moreover, differences in the shape and functions of the cortex are known contributors to the cognitive differences between sexes. Many researchers believe that sex differences in cognition are determined by biological factors, such as genetic and hormonal factors that affect brain anatomy, function, or both (7,8). The difference in hemodynamics is also presumed to be a reason for the difference in the brain volume, which may be the potential reason why cognitive performance differs between sexes (9). However, the mechanism underlying the correlation between hemodynamics and cognition requires further exploration.

Previous tissue anatomy studies have reported that typical penetrating venules appear to drain the blood supplied by 4 to 5 penetrating arterioles (10,11). A schematic representation of the vascular anatomical structure of the veins and arteries in the cortex is shown in Figure 1A. Due to these anatomical features of the cortical vasculature, stenosis, or occlusion of one penetrating venule can evidently increase resistance in multiple upstream arterioles. The mechanism underlying the hemodynamic changes due to this venous structure is shown in Figure 1B. Given the aforementioned information, we hypothesized that, because of the characteristics of the human cortical angioarchitecture, the penetrating venules located at the center of the perfusion domain may be a point of vulnerability in patients with cerebrovascular disease. Thus, an exploration of the potential relationship between cerebral veins and cerebrovascular-related diseases and the related mechanism may be valuable. It is well-known that susceptibility-weighted imaging (SWI) is the primary method for noninvasively investigating the cerebral veins in vivo. The SWI sequence uses full velocity-compensated high-resolution 3-dimensional (3D) gradient echo sequences, in which the difference in magnetic susceptibility between tissues is used to enhance the image contrast. Based on this imaging principle, SWI has been used to visualize hemosiderin, deoxyhaemoglobin, and other substances. Therefore, SWI is useful for observing the intracranial venous system. Moreover, it has gradually become a powerful and useful tool to delineate venous structures in the brain and to study diverse pathological conditions (12-15).

SCVs refer to the cerebral veins in the area of the cerebral cortex. In the MinIP images, the SCVs are represented by the linear black part of the cerebral cortex, which is distinguished from the surrounding white brain parenchyma. Our objective in this study was to identify the SCVs and automatically quantify their morphological characteristics; notably, only the brain parenchyma needed to be analyzed. Therefore, the extraction of the brain parenchyma from the image, as the outer contour of the brain, might have affected the learning of the neural network.

The experiment was performed on a Windows 10 system, and the central processing unit (CPU) was an Intel Core i7-8565U1.80 GHz (Santa Clara, CA, USA). PyTorch framework version 1.4 (Linux Foundation, San Francisco, CA, USA) and Python version 3.6 (Python Software Foundation, Wilmington, DE, USA) were used. The 1,000 sets of data were randomly divided into training (n=600), validation (n=200), and test sets (n=200). The amount of data used for establishing a deep learning model at each stage is summarized in Table 1. In the experiment, the DeepLabV3+ network (27) (Hasty, Berlin, Germany) was used to train the segmentation image of SCVs. The main body of this network structure decoder is a deep convolutional neural network (DCNN) with atrous convolution, and its basic network structure, Resnet, is used to extract image features. In addition, the network structure decoder also has atrous spatial pyramid pooling (ASPP), which is mainly used to introduce multiscale information and a decoding module and, in addition, to fuse low-level features with high-level features and improve the accuracy of segmentation boundaries. A gradual decrease method was adopted for the learning rate to avoid falling into a local optimum; specifically, after every 200 iterations, the learning rate was adjusted. The input image size was fixed to 512512, and the output data were the model classification prediction results. The main parameters for training the deep learning model were the following: model depth =19, hidden layer =16, dropout =0.5 [dropout may effectively alleviate the occurrence of overfitting and achieve the effect of regularization to a certain extent (28)], batch size =64 (batch size refers to the number of data samples captured in a training session), pretrained = true, number of epochs =200 (epochs refer to the number of times all data have been traversed), and learning rate =0.01 (the learning rate controls the learning progress of the model). The neural network structure training process is shown in Figure 3. The segmentation and recognition steps for the SCV images are shown in Figure 4.

Statistical analyses were conducted using SPSS software (version 25.0; IBM Corp., Armonk, NY, USA). Study participants were categorized by sex. Measurement data with a normal distribution are reported as means standard deviations, and the independent sample t test was used for comparisons between groups. Measurement data with a skewed distribution are presented as the median and the first and third quartile [M (Q1, Q3)], and the Mann-Whitney test was used to compare results between groups. Potential SCV-related effects on cognition were investigated by calculating the Spearman correlation coefficients, with the cognitive score and the quantified characteristics of SCVs serving as variables. A P value less than 0.05 indicated a statistically significant difference.

The results for the evaluation of the performance of the deep learning model in analyzing both the training and validation sets are shown in Figure 5. The accuracy is defined as the mean of the union of the intersection area ratios of the marker map and the recognition map. The epochs refer to the number of times all data have been traversed. Generally, the accuracy of the model increased consistently as the number of epochs increased. After 20 epochs, the model eventually stabilized and was highly accurate for both the training and validation sets, as shown by the blue curved line in Figure 5. Finally, the accuracy of the deep learning model reached 98.86% for the training set and 98.02% for the validation set. The loss values obtained in the process of model training and model validation are also shown in Figure 5 and are indicated by the green curve. The loss function decreased gradually as the number of epochs increased, indicating good model convergence.

Demographic information and cognitive performance of the volunteers in this study are shown in Table 2. A total of 144 volunteers were included in this study. The average age was 59.64 years, the numbers of males and females were 72 (50%) and 72 (50%), respectively, and the average education level was 11.49 years. No significant differences in age or education were observed between the groups. The results shown in Figure 6 indicate that in terms of cognitive performance, males scored significantly higher than females on both the MMSE and MOCA-B (P=0.016 and P=0.015, respectively).

3a8082e126