Search Results (45)

Background: Food image recognition, a crucial step in computational gastronomy, has diverse applications across nutritional platforms. Convolutional neural networks (CNNs) are widely used for this task due to their ability to capture hierarchical features. However, they struggle with long-range dependencies and global feature extraction, which are vital in distinguishing visually similar foods or images where the context of the whole dish is crucial, thus necessitating transformer architecture. Objectives: This research explores the capabilities of the CNNs and transformers to build a robust classification model that can handle both short- and long-range dependencies with global features to accurately classify food images and enhance food image recognition for better nutritional analysis. Methods: Our approach, which combines CNNs and Vision Transformers (ViTs), begins with the RestNet50 backbone model. This model is responsible for local feature extraction from the input image. The resulting feature map is then passed to the ViT encoder block, which handles further global feature extraction and classification using multi-head attention and fully connected layers with pre-trained weights. Results: Our experiments on five diverse datasets have confirmed a superior performance compared to the current state-of-the-art methods, and our combined dataset leveraging complementary features showed enhanced generalizability and robust performance in addressing global food diversity. We used explainable techniques like grad-CAM and LIME to understand how the models made their decisions, thereby enhancing the user’s trust in the proposed system. This model has been integrated into a mobile application for food recognition and nutrition analysis, offering features like an intelligent diet-tracking system. Conclusion: This research paves the way for practical applications in personalized nutrition and healthcare, showcasing the extensive potential of AI in nutritional sciences across various dietary platforms. Full article

Human pose estimation is an important research direction in the field of computer vision, which aims to accurately identify the position and posture of keypoints of the human body through images or videos. However, multi-person pose estimation yields false detection or missed detection in dense crowds, and it is still difficult to detect small targets. In this paper, we propose a Mamba-based human pose estimation. First, we design a GMamba structure to be used as a backbone network to extract human keypoints. A gating mechanism is introduced into the linear layer of Mamba, which allows the model to dynamically adjust the weights according to the different input images to locate the human keypoints more precisely. Secondly, GMamba as the backbone network can effectively solve the long-sequence problem. The direct use of convolutional downsampling reduces selectivity for different stages of information flow. We used slice downsampling (SD) to reduce the resolution of the feature map to half the original size, and then fused local features from four different locations. The fusion of multi-channel information helped the model obtain rich pose information. Finally, we introduced an adaptive threshold focus loss (ATFL) to dynamically adjust the weights of different keypoints. We assigned higher weights to error-prone keypoints to strengthen the model’s attention to these points. Thus, we effectively improved the accuracy of keypoint identification in cases of occlusion, complex background, etc., and significantly improved the overall performance of attitude estimation and anti-interference ability. Experimental results showed that the AP and AP50 of the proposed algorithm on the COCO 2017 validation set were 72.2 and 92.6. Compared with the typical algorithm, it was improved by 1.1% on AP50. The proposed method can effectively detect the keypoints of the human body, and provides stronger robustness and accuracy for the estimation of human posture in complex scenes. Full article

Background: Obesity and type 2 diabetes (T2D) promote inflammation, increasing the risk of colorectal cancer (CRC). High-fat diet (HFD)-induced obesity is key to these diseases through biological mechanisms. This study examined the impact of genetic background on the multimorbidity of intestinal cancer, T2D, and inflammation due to HFD-induced obesity. Methods: A cohort of 357 Collaborative Cross (CC) mice from 15 lines was fed either a control chow diet (CHD) or HFD for 12 weeks. Body weight was tracked biweekly, and blood glucose was assessed at weeks 6 and 12 via intraperitoneal glucose tolerance tests (IPGTT). At the study’s endpoint, intestinal polyps were counted, and cytokine profiles were analyzed to evaluate the inflammatory response. Results: HFD significantly increased blood glucose levels and body weight, with males showing higher susceptibility to T2D and obesity. Genetic variation across CC lines influenced glucose metabolism, body weight, and polyp development. Mice on HFD developed more intestinal polyps, with males showing higher counts than females. Cytokine analysis revealed diet-induced variations in pro-inflammatory markers like IL-6, IL-17A, and TNF-α, differing by genetic background and sex. Conclusions: Host genetics plays a crucial role in susceptibility to HFD-induced obesity, T2D, CRC, and inflammation. Genetic differences across CC lines contributed to variability in disease outcomes, providing insight into the genetic underpinnings of multimorbidity. This study supports gene-mapping efforts to develop personalized prevention and treatment strategies for these diseases. Full article

Brain tumor detection is crucial for effective treatment planning and improved patient outcomes. However, existing methods often face challenges, such as limited interpretability and class imbalance in medical-imaging data. This study presents a novel, custom Convolutional Neural Network (CNN) architecture, specifically designed to address these issues by incorporating interpretability techniques and strategies to mitigate class imbalance. We trained and evaluated four CNN models (proposed CNN, ResNetV2, DenseNet201, and VGG16) using a brain tumor MRI dataset, with oversampling techniques and class weighting employed during training. Our proposed CNN achieved an accuracy of 94.51%, outperforming other models in regard to precision, recall, and F1-Score. Furthermore, interpretability was enhanced through gradient-based attribution methods and saliency maps, providing valuable insights into the model’s decision-making process and fostering collaboration between AI systems and clinicians. This approach contributes a highly accurate and interpretable framework for brain tumor detection, with the potential to significantly enhance diagnostic accuracy and personalized treatment planning in neuro-oncology. Full article

As part of an investigation to detect asymmetries in gait patterns in persons with shoulder injuries, the goal of the present study was to design and validate a Kinect-based motion capture system that would enable the extraction of joint kinematics curves during gait and to compare them with the data obtained through a commercial motion capture system. The study included eight male and two female participants, all diagnosed with anterolateral shoulder pain syndrome in their right upper extremity with a minimum 18 months of disorder evolution. The participants had an average age of 31.8 ± 9.8 years, a height of 173 ± 18 cm, and a weight of 81 ± 15 kg. The gait kinematics were sampled simultaneously with the new system and the Clinical 3DMA system. Shoulder, elbow, hip, and knee kinematics were compared between systems for the pathological and non-pathological sides using repeated measures ANOVA and 1D statistical parametric mapping. For most variables, no significant difference was found between systems. Evidence of a significant difference between the newly developed system and the commercial system was found for knee flexion–extension (p < 0.004, between 60 and 80% of the gait cycle), and for shoulder abduction–adduction. The good concurrent validity of the new Kinect-based motion analysis system found in this study opens promising perspectives for clinical motion tracking using an affordable and simple system. Full article

Estimating the calorie content of fruits is critical for weight management and maintaining overall health as well as aiding individuals in making informed dietary choices. Accurate knowledge of fruit calorie content assists in crafting personalized nutrition plans and preventing obesity and associated health issues. In this paper, we investigate the application of deep learning models for estimating the calorie content in fruits from digital images, aiming to provide a more efficient and accurate method for nutritional analysis. We create a dataset comprising images of various fruits and employ random data augmentation techniques during training to enhance model robustness. We utilize the RT-DETR model integrated into the ultralytics framework for implementation and conduct comparative experiments with YOLOv10 on the dataset. Our results show that the RT-DETR model achieved a precision rate of 99.01% and mAP50-95 of 94.45% in fruit detection from digital images, outperforming YOLOv10 in terms of F1- Confidence Curves, P-R curves, precision, and mAP. Conclusively, in this paper, we utilize a transformer architecture to detect fruits and estimate their calorie and nutritional content. The results of the experiments provide a technical reference for more accurately monitoring an individual’s dietary intake by estimating the calorie content of fruits. Full article

Sarcopenia, the age-related loss of skeletal muscle mass, is a core component of frailty that is associated with functional decline and adverse health events in older adults. Unfortunately, the available tools to diagnose sarcopenia are often inaccessible or not user-friendly for clinicians. Point-of-care ultrasound (US) is a promising tool that has been used to image the quadriceps muscle and measure its thickness (QMT) as a diagnostic criterion for sarcopenia. This measurement can be challenging for clinicians, especially when performed at the bedside using handheld systems or phased-array probes not designed for this use case. In this paper, we sought to automate this measurement using deep learning methods to improve its accuracy, reliability, and speed in the hands of untrained clinicians. In the proposed framework, which aids in better training, particularly when limited data are available, convolutional and transformer-based deep learning models with generic or data-driven pre-trained weights were compared. We evaluated regression (QMT as a continuous output in cm) and classification (QMT as an ordinal output in 0.5 cm bins) approaches, and in the latter, activation maps were generated to interpret the anatomical landmarks driving the model predictions. Finally, we evaluated a segmentation approach to derive QMT. The results showed that both transformer-based models and convolutional neural networks benefit from the proposed framework in estimating QMT. Additionally, the activation maps highlighted the interface between the femur bone and the quadriceps muscle as a key anatomical landmark for accurate predictions. The proposed framework is a pivotal step to enable the application of US-based measurement of QMT in large-scale clinical studies seeking to validate its diagnostic performance for sarcopenia, alone or with ancillary criteria assessing muscle quality or strength. We believe that implementing the proposed framework will empower clinicians to conveniently diagnose sarcopenia in clinical settings and accordingly personalize the care of older patients, leading to improved patient outcomes and a more efficient allocation of healthcare resources. Full article

The virtual try-on technology based on 2D images aims to seamlessly transfer provided garments onto target person images. Prior methods mainly concentrated on warping garments and generating images, overlooking the influence of feature alignment on the try-on results. In this study, we initially analyze the distortions present by existing methods and elucidate the critical role of feature alignment in the extraction stage. Building on this, we propose a novel feature alignment-based model (FA-VTON). Specifically, FA-VTON aligns the upsampled higher-level features from both person and garment images to acquire precise boundary information, which serves as guidance for subsequent garment warping. Concurrently, the Efficient Channel Attention mechanism (ECA) is introduced to generate the final result in the try-on generation module. This mechanism enables adaptive adjustment of channel feature weights to extract important features and reduce artifact generation. Furthermore, to make the student network focus on salient regions of each channel, we utilize channel-wise distillation (CWD) to minimize the Kullback–Leibler (KL) divergence between the channel probability maps of the two networks. The experiments show that our model achieves better results in both qualitative and quantitative analyses compared to current methods on the popular virtual try-on datasets. Full article

This study introduces a data-driven and machine-learning approach to design a personalized tourist recommendation system for Nepal. It examines key tourist attributes, such as demographics, behaviors, preferences, and satisfaction, to develop four sub-models for data collection and machine learning. A structured survey is conducted with 2400 international and domestic tourists, featuring 28 major questions and 125 variables. The data are preprocessed, and significant features are extracted to enhance the accuracy and efficiency of the machine-learning models. These models are evaluated using metrics such as accuracy, precision, recall, F-score, ROC, and lift curves. A comprehensive database for Pokhara City, Nepal, is developed from various sources that includes attributes such as location, cost, popularity, rating, ranking, and trend. The machine-learning models provide intermediate categorical recommendations, which are further mapped using a personalized recommender algorithm. This algorithm makes decisions based on weights assigned to each decision attribute to make the final recommendations. The system’s performance is compared with other popular recommender systems implemented by TripAdvisor, Google Maps, the Nepal tourism website, and others. It is found that the proposed system surpasses existing ones, offering more accurate and optimized recommendations to visitors in Pokhara. This study is a pioneering one and holds significant implications for the tourism industry and the governing sector of Nepal in enhancing the overall tourism business. Full article

Background: The tumor histological grade is closely related to the prognosis of patients with endometrial cancer (EC). Multiparametric MRI, including diffusion-weighted imaging (DWI), provides information about the cellular density that may be useful to differentiate between benign and malignant uterine lesions. However, correlations between apparent diffusion coefficient (ADC) values and histopathological grading in endometrial cancer remain controversial. Material and methods: We retrospectively evaluated 92 patients with endometrial cancers, including both endometrioid adenocarcinomas (64) and non-endometrioid adenocarcinomas (28). All patients underwent DWI procedures, and mean ADC values were calculated in a region of interest. These values were then correlated with the tumor grading offered by the histopathological examination, which was considered the gold standard. In this way, the patients were divided into three groups (G1, G2, and G3). The ADC values were then compared to the results offered by the biopsy to see if the DWI sequence and ADC map could replace this procedure. We also compared the mean ADC values to the myometrial invasion (</>50%) and lymphovascular space invasion. Results: We have divided the ADC values into three categories corresponding to three grades: >0.850 × 10⁻³ mm²/s (ADC1), 0.730–0.849 × 10⁻³ mm²/s (ADC2) and <0.730 × 10⁻³ mm²/s (ADC3). The diagnostic accuracy of the ADC value was 85.71% for ADC1, 75.76% for ADC2, and 91.66% for ADC3. In 77 cases out of 92, the category in which they were placed using the ADC value corresponded to the result offered by the histopathological exam with an accuracy of 83.69%. For only 56.52% of patients, the biopsy result included the grading system. For each grading category, the mean ADC value showed better results than the biopsy; for G1 patients, the mean ADC value had an accuracy of 85.71% compared to 66.66% in the biopsy, G2 had 75.76% compared to 68.42%, and G3 had 91.66 compared to 75%. For both deep myometrial invasion and lymphovascular space invasion, there is a close, inversely proportional correlation with the mean ADC value. Conclusions: Mean endometrial tumor ADC on MR-DWI is inversely related to the histological grade, deep myometrial invasion and lymphovascular space invasion. Using this method, the patients could be better divided into risk categories for personalized treatment. Full article

Type 2 diabetes mellitus (T2DM) poses a higher risk for complications in South Asian individuals compared to other ethnic groups. To shed light on potential mediating factors, we investigated lipidomic changes in plasma of Dutch South Asians (DSA) and Dutch white Caucasians (DwC) with and without T2DM and explore their associations with clinical features. Using a targeted quantitative lipidomics platform, monitoring over 1000 lipids across 17 classes, along with ¹H NMR based lipoprotein analysis, we studied 51 healthy participants (21 DSA, 30 DwC) and 92 T2DM patients (47 DSA, 45 DwC) from the MAGNetic resonance Assessment of VICTOza efficacy in the Regression of cardiovascular dysfunction in type 2 dIAbetes mellitus (MAGNA VICTORIA) study. This comprehensive mapping of the circulating lipidome allowed us to identify relevant lipid modules through unbiased weighted correlation network analysis, as well as disease and ethnicity related key mediatory lipids. Significant differences in lipidomic profiles, encompassing various lipid classes and species, were observed between T2DM patients and healthy controls in both the DSA and DwC populations. Our analyses revealed that healthy DSA, but not DwC, controls already exhibited a lipid profile prone to develop T2DM. Particularly, in DSA-T2DM patients, specific lipid changes correlated with clinical features, particularly diacylglycerols (DGs), showing significant associations with glycemic control and renal function. Our findings highlight an ethnic distinction in lipid modules influencing clinical outcomes in renal health. We discover distinctive ethnic disparities of the circulating lipidome and identify ethnicity-specific lipid markers. Jointly, our discoveries show great potential as personalized biomarkers for the assessment of glycemic control and renal function in DSA-T2DM individuals. Full article

Humanlike driving is significant in improving the safety and comfort of automated vehicles. This paper proposes a personalized motion planning method with driver characteristics in longitudinal and lateral directions for highway automated driving. The motion planning is decoupled into path optimization and speed optimization under the framework of the Baidu Apollo EM motion planner. For modeling driver behavior in the longitudinal direction, a car-following model is developed and integrated into the speed optimizer based on a weight ratio hypothesis model of the objective functional, whose parameters are obtained by Bayesian optimization and leave-one-out cross validation using the driving data. For modeling driver behavior in the lateral direction, a Bayesian network (BN), which maps the physical states of the ego vehicle and surrounding vehicles and the lateral intentions of the surrounding vehicles to the driver’s lateral intentions, is built in an efficient and lightweight way using driving data. Further, a personalized reference trajectory decider is developed based on the BN, considering traffic regulations, the driver’s preference, and the costs of the trajectories. According to the actual traffic scenarios in the driving data, a simulation is constructed, and the results validate the human likeness of the proposed motion planning method. Full article

(1) Background: Colorectal cancer is the third most common type of cancer with a high mortality rate and poor prognosis. The accurate prediction of key genetic mutations, such as the KRAS status, tumor staging, and extramural venous invasion (EMVI), is crucial for guiding personalized treatment decisions and improving patients’ outcomes. MRI radiomics was assessed to predict the KRAS status and tumor staging in colorectal cancer patients across different imaging platforms to improve the personalized treatment decisions and outcomes. (2) Methods: Sixty colorectal cancer patients (35M/25F; avg. age 56.3 ± 12.9 years) were treated at an oncology unit. The MRI scans included T2-weighted (T2W) and diffusion-weighted imaging (DWI) or the apparent diffusion coefficient (ADC). The manual segmentation of colorectal cancer was conducted on the T2W and DWI/ADC images. The cohort was split into training and validation sets, and machine learning was used to build predictive models. (3) Results: The neural network (NN) model achieved 73% accuracy and an AUC of 0.71 during training for predicting the KRAS mutation status, while during testing, it achieved 62.5% accuracy and an AUC of 0.68. In the case of tumor grading, the support vector machine (SVM) model excelled with a training accuracy of 72.93% and an AUC of 0.7, and during testing, it reached an accuracy of 72% and an AUC of 0.69. (4) Conclusions: ML models using radiomics from ADC maps and T2-weighted images are effective for distinguishing KRAS genes, tumor grading, and EMVI in colorectal cancer. Standardized protocols are essential to improve MRI radiomics’ reliability in clinical practice. Full article

High tibial osteotomy (HTO) is intended to treat medial knee osteoarthritis by realigning the joint such that the loading in the knee during functional activity shifts laterally. The aim of this study was to use a novel methodology combining motion analysis and 3D modelling to assess the efficacy of this surgery in changing the loading location in the knee in a cohort of 25 patients treated with personalized HTO. Pre-operatively and at 6 months post-surgery, weight-bearing CT and gait analysis during level walking were performed on all patients, as well as clinical evaluations using KOOS and VAS scores. CT scans were used to generate a knee bone model and a virtual tibial plateau plane; the intersection pattern between this plane and the ground reaction force (GRF) vector was calculated in the pre- and post-operative gait analyses. Clinical scores improved significantly (p < 0.001) after surgery (pre-/post-operative KOOS and VAS: 56.2 ± 14.0/82.0 ± 8.3 and 6.3 ± 1.7/1.5 ± 1.7). Post-operative GRF-to-tibial plateau intersection patterns were significantly (p < 0.001) more lateral (31.9 ± 19.8% of tibial plateau width) than the pre-operative patterns. Personalized HTO successfully and consistently lateralizes the GRF at the knee, in association with significant improvements in function and pain. The novel combination of 3D bone modelling and motion analysis also has the potential to further aid HTO surgical planning. Full article

Occluded pedestrian detection faces huge challenges. False positives and false negatives in crowd occlusion scenes will reduce the accuracy of occluded pedestrian detection. To overcome this problem, we proposed an improved you-only-look-once version 3 (YOLOv3) based on squeeze-and-excitation networks (SENet) and optimized generalized intersection over union (GIoU) loss for occluded pedestrian detection, namely YOLOv3-Occlusion (YOLOv3-Occ). The proposed network model considered incorporating squeeze-and-excitation networks (SENet) into YOLOv3, which assigned greater weights to the features of unobstructed parts of pedestrians to solve the problem of feature extraction against unsheltered parts. For the loss function, a new generalized intersection over union_{intersection over groundtruth} (GIoU_IoG) loss was developed to ensure the areas of predicted frames of pedestrian invariant based on the GIoU loss, which tackled the problem of inaccurate positioning of pedestrians. The proposed method, YOLOv3-Occ, was validated on the CityPersons and COCO2014 datasets. Experimental results show the proposed method could obtain 1.2% MR⁻² gains on the CityPersons dataset and 0.7% mAP@50 improvements on the COCO2014 dataset. Full article