Using deep learning to predict ideology from facial photographs: Expressions, beauty, and extra-facial information

Deep learning techniques can use common public data such as facial photographs to predict sensitive personal information, but little is known about what information contributes to the predictive success of these techniques. This lack of knowledge limits both the public’s ability to protect against revealing unintended information as well as the scientific utility of deep learning results. We combine convolutional neural networks, heat maps, facial expression coding, and classification of identifiable features such as masculinity and attractiveness in our study of political ideology in 3323 Danes. Predictive accuracy from the neural network was 61% in each gender. Model-predicted ideology correlated with aspects of both facial expressions (happiness vs neutrality) and morphology (specifically, attractiveness in females). Heat maps highlighted the informativeness of areas both on and off the face, pointing to methodological refinements and the need for future research to better understand the significance of certain facial areas.

Explainable AI for CNN-based Prostate Tumor Segmentation in Multi-parametric MRI Correlated to Whole Mount Histopathology

Automatic prostate tumor segmentation is often unable to identify the lesion even if in multi-parametric MRI data is used as input, and the segmentation output is difficult to verify due to the lack of clinically established ground truth images. In this work we use an explainable deep learning model to interpret the predictions of a convolutional neural network (CNN) for prostate tumor segmentation. The CNN uses a U-Net architecture which was trained on multi-parametric MRI data from 122 patients to automatically segment the prostate gland and prostate tumor lesions. In addition, co-registered ground truth data from whole mount histopathology images were available in 15 patients that were used as a test set during CNN testing. To be able to interpret the segmentation results of the CNN, heat maps were generated using the Gradient Weighted Class Activation Map (Grad-CAM) method. With the CNN a mean Dice Sorensen Coefficient for the prostate gland and the tumor lesions of 0.62 and 0.31 with the radiologist drawn ground truth and 0.32 with wholemount histology ground truth for tumor lesions could be achieved. Dice Sorensen Coefficient between CNN predictions and manual segmentations from MRI and histology data were not significantly different. In the prostate the Grad-CAM heat maps could differentiate between tumor and healthy prostate tissue, which indicates that the image information in the tumor was essential for the CNN segmentation.

Usability analysis taking into consideration the aspects of accessibility of selected university websites

The goal of this paper was to evaluate selected web services of universities in terms of user experience, with particular emphasis on usability and accessibility. The research was conducted using eye-tracking and questionnaire methods. Ten people participated in this study. The objects of the study were three university websites: the Catholic University of Lublin (KUL), the Cracow University of Technology (PK) and the West Pomeranian University of Technology (ZUT). The eye-tracking data were subjected to qualitative and quantitative analyses, while the data from questionnaires were subjected to quantitative analysis. The results of individual analyses are presented in the form of heat maps, scan paths, charts and tables.

Correlation Evaluation of Pilots’ Situation Awareness in Bridge Simulations via Eye-Tracking Technology

To maintain situation awareness (SA) when exposed to emergencies during pilotage, a pilot needs to selectively allocate attentional resources to perceive critical status information about ships and environments. Although it is important to continuously monitor a pilot’s SA, its relationship with attention is still not fully understood in ship pilotage. This study performs bridge simulation experiments that include vessel departure, navigation in the fairway, encounters, poor visibility, and anchoring scenes with 13 pilots (mean = 11.3 and standard deviation = 1.4 of experience). Individuals were divided into two SA group levels based on the Situation Awareness Rating Technology (SART-2) score (mean = 20.13 and standard deviation = 5.83) after the experiments. The visual patterns using different SA groups were examined using heat maps and scan paths based on pilots’ fixations and saccade data. The preliminary visual analyses of the heat maps and scan paths indicate that the pilots’ attentional distribution is modulated by the SA level. That is, the most concerning areas of interest (AOIs) for pilots in the high and low SA groups are outside the window (AOI-2) and electronic charts (AOI-1), respectively. Subsequently, permutation simulations were utilized to identify statistical differences between the pilots’ eye-tracking metrics and SA. The results of the statistical analyses show that the fixation and saccade metrics are affected by the SA level in different AOIs across the five scenes, which confirms the findings of previous studies. In encounter scenes, the pilots’ SA level is correlated with the fixation and saccade metrics: fixation count ( p = 0.034 < 0.05 in AOI-1 and p = 0.032 < 0.05 in AOI-2), fixation duration ( p = 0.043 < 0.05 in AOI-1 and p = 0.014 < 0.05 in AOI-2), and saccade count ( p = 0.086 < 0.1 in AOI-1 and p = 0.054 < 0.1 in AOI-2). This was determined by the fixation count ( p = 0.024 < 0.05 in AOI-1 and p = 0.034 < 0.05 in AOI-2), fixation duration ( p = 0.036 < 0.05 in AOI-1 and p = 0.047 < 0.05 in AOI-2), and saccade duration ( p = 0.05 ≤ 0.05 in AOI-1 and p = 0.042 < 0.05 in AOI-2) in poor-visibility scenes. In the remaining scenes, the SA could not be measured using eye movements alone. This study lays a foundation for the cognitive mechanism recognition of pilots based on SA via eye-tracking technology, which provides a reference to establish cognitive competency standards in preliminary pilot screenings.

Artificial Intelligence Assistive Technology in Hospital Professional Nursing Technology

Global aging is becoming more and more serious, and the nursing problems of the elderly will become very serious in the future. The article designs a control system with ATmega128 as the main controller based on the function of the multifunctional nursing robot. The article uses a convolutional neural network structure to estimate the position of 3D human joints. The article maps the joint coordinates of the colour map to the depth map based on the two camera parameters. At the same time, 15 joint heat maps are constructed with the joint depth map coordinates as the centre, and the joint heat map and the depth map are bound to the second-level neural network. The prediction of the position of the user’s armpit is further completed by image processing technology. We compare this method with other attitude prediction methods to verify the advantages of this research method. The research background of this article is carried out in the context of global aging in the 21st century.

Eye tracking and tourism research: A systematic literature review

In recent years, the number of studies in tourism using the eye tracking technique has increased and started generating valuable information for both academics and the industry. However, there is a gap in the literature concerning systematic reviews focused on recent articles and their findings. Thus, the aim of this study is to close this gap by systematically analysing 70 research papers tackling the subject of eye tracking in tourism and published in highly ranked tourism journals. The study identifies the most popular topics and trends for eye tracking research, as well as the most used types of visual stimuli, such as exhibitions, restaurant menus, promotional pictures or websites. The study also details on measurements specific for the analysis of eye tracking data, including fixations, saccades and heat maps. Results are emphasized along with their theoretical and practical implications. In addition, we highlight the lack of the use of dynamic stimuli in the existing literature and suggest further research directions using the eye tracking technique.

Abstract 1122‐000188: 3D Analysis of Atherosclerotic Plaque Enhancement and the Parent Vessel

Introduction : High resolution vessel wall imaging (HR‐VWI) is a promising tool in studying intracerebral atherosclerotic disease. The analysis of the interplay between the patterns of enhancement between the plaque and its parent vessel can generate further insights on the biology of these lesions. We have developed a 3D method of plaque and parent vessel analysis. Methods : Images from fifty‐five plaques were obtained using 7T HR‐VWI. T1 and T1+Gd sequences were performed. 3D reconstructions of the plaque and its parent vessel were generated with 3D Slicer. Using an in‐house code, probes were orthogonally extended from the lumen of the vessel into the vessel wall and the plaque. Signal intensity values were then normalized to the corpus callosum. 3D heat maps and histograms were generated from hundreds of data points. A detailed analysis of the morphology of the histograms was performed to determine the uptake of gadolinium (Gd) by the parent vessel and the plaque. Variations in the width of the histogram were measured with the standard deviation. Results : Forty‐one patients with 55 plaques (41 culprit and 15 non culprit) were included. There was no difference in enhancement in T1‐pre between culprit and non‐culprit plaques when compared to the parent vessel (width = 0.14 ± 0.05 and 0.14 ± 0.03, respectively; p = 0.91). On the T1+Gd culprit plaques were significantly more enhancing compared to the parent vessel (0.26 ± 0.10) than non‐culprit plaques (0.20 ± 0.06) (p = 0.02). The presence of an enhancing plaque creates a bimodal distribution that increases the width of the histogram curve (figure). Conclusions : Culprit plaques exhibit different patterns of enhancement relative to the parent vessel compared to non‐culprit plaques. Histogram analysis of the parent vessel and its plaques provides a new set of metrics that may be used as a biomarker of disease progression.