Ghazal Rouhafzay
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ghazal Rouhafzay has a total of 47 co-authors. Here are the top co-authors by number of shared publications:
- Ana-Maria Cretu
Université du Québec en Outaouais
7 shared publications Ana-Maria Cretu
Department of Engineering and Computer Science, Université du Québec en Outaouais, Gatineau, Canada
6 shared publications- Jalila Jbilou
Université de Moncton
4 shared publications Pierre Payeur
University of Ottawa,Ottawa,ON,Canada
4 shared publications- Hajarimino Rakotomanana
Department of Computer Science, Université de Moncton, Moncton, NB E1A 3E9, Canada
3 shared publications - Abdelatif Guerroudj
Centre de formation médicale du Nouveau-Brunswick, Moncton, CA
2 shared publications
Latest Publications
- Open Access
- 30 October 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 163 (1) , 63
FusionPhys: A Multi-Source rPPG Method for Estimating Heart Rate from Facial Videos
Remote photoplethysmography (rPPG) estimates cardiovascular pulse information from facial videos without the use of contact sensors. Considerable research effort has been devoted over the past decade to improve the accuracy of this technology using both classical color-based estimators and learned models. However, most of them remain vulnerable in real-world conditions, including subject movements, lighting variations, or low image quality. To overcome these limitations, we propose FusionPhys, an architecture that exploits three complementary sources of information extracted from facial video: the raw appearance of the face, motion variations between successive frames, and pulse indices provided by classical estimators. The model is trained to predict the standardized first derivative of PPG, from which heart rate is estimated by the dominant spectral peak. Two subject-independent evaluations were performed on UBFC-rPPG. On Dataset 1, leave-one-subject-out evaluation gave heart-rate MAEs of 1.02 BPM for POS, 1.38 BPM for FusionPhys, 2.60 BPM for DeepPhys, and 6.60 BPM for Green. On the larger Dataset 2, using 30/6/6 train/validation/test subjects, MAEs were 1.93, 2.37, 14.17, and 18.45 BPM, respectively. FusionPhys therefore substantially improved upon DeepPhys while remaining close to the strongest classical baseline, and it provided the lowest waveform-derivative RMSE among the evaluated methods.
- 30 October 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 163 (1) , 63
Open AccessFusionPhys: A Multi-Source rPPG Method for Estimating Heart Rate from Facial Videos
Remote photoplethysmography (rPPG) estimates cardiovascular pulse information from facial videos without the use of contact sensors. Considerable research effort has been devoted over the past decade to improve the accuracy of this technology using both classical color-based estimators and learned models. However, most of them remain vulnerable in real-world conditions, including subject movements, lighting variations, or low image quality. To overcome these limitations, we propose FusionPhys, an architecture that exploits three complementary sources of information extracted from facial video: the raw appearance of the face, motion variations between successive frames, and pulse indices provided by classical estimators. The model is trained to predict the standardized first derivative of PPG, from which heart rate is estimated by the dominant spectral peak. Two subject-independent evaluations were performed on UBFC-rPPG. On Dataset 1, leave-one-subject-out evaluation gave heart-rate MAEs of 1.02 BPM for POS, 1.38 BPM for FusionPhys, 2.60 BPM for DeepPhys, and 6.60 BPM for Green. On the larger Dataset 2, using 30/6/6 train/validation/test subjects, MAEs were 1.93, 2.37, 14.17, and 18.45 BPM, respectively. FusionPhys therefore substantially improved upon DeepPhys while remaining close to the strongest classical baseline, and it provided the lowest waveform-derivative RMSE among the evaluated methods.
- Open Access
- 14 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (20) , 4722
Artificial Intelligence for Movement Analysis in Autism Spectrum Disorder: A Scoping Review of Gait, Stereotyped Motor Movements, and Motor Behaviors
Motor differences are increasingly recognized as relevant early markers of autism spectrum disorder (ASD), although they are not part of the core diagnostic criteria. Recent advances in computer vision, pose estimation, and machine learning have enabled the quantitative analysis of movement patterns, particularly stereotyped motor movements (SMMs) and gait deviations. This scoping review examines recent artificial intelligence approaches for analyzing motor and movement patterns in individuals with ASD, with particular emphasis on stereotyped motor movements (SMMs) and gait-related differences. The reviewed studies used 2D and 3D pose estimation, conventional machine learning algorithms, and deep learning models to extract movement features from video, skeletal, depth, and sensor-based data. The included studies addressed different outcomes, including ASD versus TD classification, SMM and repetitive-behavior detection, gait abnormality classification or characterization, motor phenotype assessment, and movement measurement or validation. Because these studies examined different tasks and used different performance metrics, their results cannot be directly compared. However, small or private datasets, heterogeneous protocols, and limited external validation restrict cross-study comparison and clinical generalizability. Overall, the current evidence supports technical movement analysis rather than validated clinical ASD screening or diagnosis.
- 14 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (20) , 4722
Open AccessArtificial Intelligence for Movement Analysis in Autism Spectrum Disorder: A Scoping Review of Gait, Stereotyped Motor Movements, and Motor Behaviors
Motor differences are increasingly recognized as relevant early markers of autism spectrum disorder (ASD), although they are not part of the core diagnostic criteria. Recent advances in computer vision, pose estimation, and machine learning have enabled the quantitative analysis of movement patterns, particularly stereotyped motor movements (SMMs) and gait deviations. This scoping review examines recent artificial intelligence approaches for analyzing motor and movement patterns in individuals with ASD, with particular emphasis on stereotyped motor movements (SMMs) and gait-related differences. The reviewed studies used 2D and 3D pose estimation, conventional machine learning algorithms, and deep learning models to extract movement features from video, skeletal, depth, and sensor-based data. The included studies addressed different outcomes, including ASD versus TD classification, SMM and repetitive-behavior detection, gait abnormality classification or characterization, motor phenotype assessment, and movement measurement or validation. Because these studies examined different tasks and used different performance metrics, their results cannot be directly compared. However, small or private datasets, heterogeneous protocols, and limited external validation restrict cross-study comparison and clinical generalizability. Overall, the current evidence supports technical movement analysis rather than validated clinical ASD screening or diagnosis.
- Open Access
- 15 April 2026
- Published by MDPI AG in Preprints.org
Contrastive Representation Learning for Voice-Based Autistic Trait Identification
Early identification of Autism Spectrum Disorder (ASD) traits in infants is crucial for early intervention, which can greatly improve the child’s quality of life. Solutions that use voice analysis offer a promising non-invasive way to detect ASD. However, most current studies depend on extracting specific voice markers from certain datasets and do not include validation across different groups. In this paper, we propose a supervised contrastive learning method for identifying ASD based on infant vocalizations. We extend the Time-Frequency Consistency (TF-C) framework from self-supervised learning to a contrastive approach that uses labels. Our method takes advantage of both time-related and frequency-related data through a dual-branch encoder. It applies supervised contrastive constraints during pre-training to reduce variation within classes while boosting separation between different classes in the embedding space. We pre-train the model using diagnostic labels on a dataset that includes typically developing (TD), Attention-Deficit Hyperactivity Disorder (ADHD), and ASD infants from an open-access dataset, and then fine-tune it with a simple classification head. Evaluation on a cross-cohort group of participants shows the model generalizes well and can distinguish ASD from non-ASD infants, achieving up to 100.00 % accuracy on non-verbal vocalizations.
- 15 April 2026
- Published by MDPI AG in Preprints.org
Open AccessContrastive Representation Learning for Voice-Based Autistic Trait Identification
Early identification of Autism Spectrum Disorder (ASD) traits in infants is crucial for early intervention, which can greatly improve the child’s quality of life. Solutions that use voice analysis offer a promising non-invasive way to detect ASD. However, most current studies depend on extracting specific voice markers from certain datasets and do not include validation across different groups. In this paper, we propose a supervised contrastive learning method for identifying ASD based on infant vocalizations. We extend the Time-Frequency Consistency (TF-C) framework from self-supervised learning to a contrastive approach that uses labels. Our method takes advantage of both time-related and frequency-related data through a dual-branch encoder. It applies supervised contrastive constraints during pre-training to reduce variation within classes while boosting separation between different classes in the embedding space. We pre-train the model using diagnostic labels on a dataset that includes typically developing (TD), Attention-Deficit Hyperactivity Disorder (ADHD), and ASD infants from an open-access dataset, and then fine-tune it with a simple classification head. Evaluation on a cross-cohort group of participants shows the model generalizes well and can distinguish ASD from non-ASD infants, achieving up to 100.00 % accuracy on non-verbal vocalizations.
- Open Access
- 1 April 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14, e64139
Calorie-Counting Apps for Monitoring and Managing Calorie Intake in Adults Living With Weight-Related Chronic Diseases: Decade-Long Scoping Review (2013-2024)
Background: Overweight and obesity, as defined by the World Health Organization, correspond to BMI values of 25-29.9 kg/m² for overweight and ≥30 kg/m² for obesity. Both conditions remain major public health challenges worldwide due to their strong link with type 2 diabetes, cardiovascular disease, and hypertension, which place a heavy clinical and economic burden on health care systems. In Canada, obesity rates are notably high, with vulnerable populations disproportionately affected due to socioeconomic barriers, limited access to preventive care, and higher comorbidity rates. Calorie-counting Mobile health (mHealth) apps support dietary self-monitoring and weight control; however, varied designs and evidence complicate assessment of feasibility and effectiveness. Objective: This study aimed to systematically evaluate the structure and content of 46 calorie-counting apps, identify factors related to their acceptability and feasibility among adults living with obesity or weight-related chronic diseases, and formulate evidence-based recommendations for app developers, clinicians, and researchers. Methods: We conducted a scoping review of papers on calorie-counting apps published between January 2013 and March 2024. We identified 771 records and applied the PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) eligibility criteria. Data on app functions, features, user engagement, and acceptability and feasibility among adults with overweight or related chronic conditions were synthesized to generate practical recommendations for designing and clinically implementing calorie-counting apps. Results: A total of 68 studies met the inclusion criteria. Randomized controlled trials (23/68, 34%) and cohort studies (16/68, 24%) were the most common designs. Most studies targeted adults with overweight or obesity (53/68, 78%), while diabetes and hypertension were less frequently represented. In total, 46 distinct calorie-counting apps were identified, with MyFitnessPal and Lose It! being the most frequently studied. Nearly all apps (45/46, 98%) offered calorie logging, often through manual entry supported by food databases, and about half included goal-setting features. The most cited acceptability factors were personalization, automation, user-friendly design, and data sharing with health care professionals; barriers included technical issues, limited food databases, and manual entry. Adherence declined over time. For example, self-monitoring with MyFitnessPal decreased from 5.4 to 1.4 days per week from weeks 4 to 12, while use of Lose It! dropped to 4 days per week by the end of 12 weeks. Twelve recommendations were developed to enhance the feasibility and acceptability of calorie-counting apps for people living with weight-related chronic diseases. Conclusions: Calorie-counting apps hold potential as tools for supporting individuals living with obesity and weight-related chronic diseases. To improve clinical usability, app developers should enhance engagement via personalization and automation, ensuring food database comprehensiveness, and minimizing tracking effort. Further research should validate effectiveness and strategies for sustaining adherence, thereby informing development of user-friendly mHealth interventions.
- 1 April 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14, e64139
Open AccessCalorie-Counting Apps for Monitoring and Managing Calorie Intake in Adults Living With Weight-Related Chronic Diseases: Decade-Long Scoping Review (2013-2024)
Background: Overweight and obesity, as defined by the World Health Organization, correspond to BMI values of 25-29.9 kg/m² for overweight and ≥30 kg/m² for obesity. Both conditions remain major public health challenges worldwide due to their strong link with type 2 diabetes, cardiovascular disease, and hypertension, which place a heavy clinical and economic burden on health care systems. In Canada, obesity rates are notably high, with vulnerable populations disproportionately affected due to socioeconomic barriers, limited access to preventive care, and higher comorbidity rates. Calorie-counting Mobile health (mHealth) apps support dietary self-monitoring and weight control; however, varied designs and evidence complicate assessment of feasibility and effectiveness. Objective: This study aimed to systematically evaluate the structure and content of 46 calorie-counting apps, identify factors related to their acceptability and feasibility among adults living with obesity or weight-related chronic diseases, and formulate evidence-based recommendations for app developers, clinicians, and researchers. Methods: We conducted a scoping review of papers on calorie-counting apps published between January 2013 and March 2024. We identified 771 records and applied the PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) eligibility criteria. Data on app functions, features, user engagement, and acceptability and feasibility among adults with overweight or related chronic conditions were synthesized to generate practical recommendations for designing and clinically implementing calorie-counting apps. Results: A total of 68 studies met the inclusion criteria. Randomized controlled trials (23/68, 34%) and cohort studies (16/68, 24%) were the most common designs. Most studies targeted adults with overweight or obesity (53/68, 78%), while diabetes and hypertension were less frequently represented. In total, 46 distinct calorie-counting apps were identified, with MyFitnessPal and Lose It! being the most frequently studied. Nearly all apps (45/46, 98%) offered calorie logging, often through manual entry supported by food databases, and about half included goal-setting features. The most cited acceptability factors were personalization, automation, user-friendly design, and data sharing with health care professionals; barriers included technical issues, limited food databases, and manual entry. Adherence declined over time. For example, self-monitoring with MyFitnessPal decreased from 5.4 to 1.4 days per week from weeks 4 to 12, while use of Lose It! dropped to 4 days per week by the end of 12 weeks. Twelve recommendations were developed to enhance the feasibility and acceptability of calorie-counting apps for people living with weight-related chronic diseases. Conclusions: Calorie-counting apps hold potential as tools for supporting individuals living with obesity and weight-related chronic diseases. To improve clinical usability, app developers should enhance engagement via personalization and automation, ensuring food database comprehensiveness, and minimizing tracking effort. Further research should validate effectiveness and strategies for sustaining adherence, thereby informing development of user-friendly mHealth interventions.
- Open Access
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
ViCo-LWIR: Vision-Based detection of Indoor Liquid Spills Using Long-Wave Infrared Imaging and a Weakly Supervised Contrastive Framework
Liquid spill detection is a critical concern in safetysensitive indoor environments, such as industrial facilities. When a spill goes unnoticed, it can result in serious consequences, including damage to equipment or products, disruption of operations, and increased safety risks. Traditional cameras that rely on visible light often struggle to detect spills effectively. This is due to the fact that many liquids are clear or semi-transparent, spread in unpredictable patterns, blend into surrounding surfaces, or are difficult to detect under varying lighting conditions. In this paper, we propose a robust liquid detection system using longwave infrared (LWIR) imaging. Despite capable instrumentation for sensing liquid spills, developing a reliable supervised learning system remains a significant challenge. Liquid spills often have unclear boundaries; their thermal behavior can vary depending on the surface, and collecting the detailed pixel-level annotations needed for training detection and segmentation models is both time-consuming and expensive. To address these challenges, we propose a weakly supervised learning framework within a contrastive learning setup. This approach significantly reduces the need for detailed annotations by learning meaningful representations of samples that contain liquid. Our approach remains robust across different surfaces and environmental conditions by explicitly accounting for the uncertainty associated with liquid spills. This objective is achieved by forcing the model to learn grouping semantically similar samples, even when their visual appearances vary through an uncertainty-aware weighting scheme. Samples that do not contain liquid are identified using a hard negative mining strategy. Additionally, training stability and convergence are improved through a curriculum learning scheme that gradually increases the difficulty of the task. Experimental results and comparisons with other weakly supervised methods confirm the robustness and generalization capability of the proposed approach in challenging detection scenarios.
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessViCo-LWIR: Vision-Based detection of Indoor Liquid Spills Using Long-Wave Infrared Imaging and a Weakly Supervised Contrastive Framework
Liquid spill detection is a critical concern in safetysensitive indoor environments, such as industrial facilities. When a spill goes unnoticed, it can result in serious consequences, including damage to equipment or products, disruption of operations, and increased safety risks. Traditional cameras that rely on visible light often struggle to detect spills effectively. This is due to the fact that many liquids are clear or semi-transparent, spread in unpredictable patterns, blend into surrounding surfaces, or are difficult to detect under varying lighting conditions. In this paper, we propose a robust liquid detection system using longwave infrared (LWIR) imaging. Despite capable instrumentation for sensing liquid spills, developing a reliable supervised learning system remains a significant challenge. Liquid spills often have unclear boundaries; their thermal behavior can vary depending on the surface, and collecting the detailed pixel-level annotations needed for training detection and segmentation models is both time-consuming and expensive. To address these challenges, we propose a weakly supervised learning framework within a contrastive learning setup. This approach significantly reduces the need for detailed annotations by learning meaningful representations of samples that contain liquid. Our approach remains robust across different surfaces and environmental conditions by explicitly accounting for the uncertainty associated with liquid spills. This objective is achieved by forcing the model to learn grouping semantically similar samples, even when their visual appearances vary through an uncertainty-aware weighting scheme. Samples that do not contain liquid are identified using a hard negative mining strategy. Additionally, training stability and convergence are improved through a curriculum learning scheme that gradually increases the difficulty of the task. Experimental results and comparisons with other weakly supervised methods confirm the robustness and generalization capability of the proposed approach in challenging detection scenarios.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Canadian Journal of Electrical and Computer Engineering
- Vol. 49 (4) , 757-768
ViCo-LWIR: Vision-Based Detection of Indoor Liquid Spills Using Long-Wave Infrared Imaging and a Weakly Supervised Contrastive Framework
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Canadian Journal of Electrical and Computer Engineering
- Vol. 49 (4) , 757-768
ViCo-LWIR: Vision-Based Detection of Indoor Liquid Spills Using Long-Wave Infrared Imaging and a Weakly Supervised Contrastive Framework
- Open Access
- 7 November 2025
- Published by MDPI AG in Engineering Proceedings
- Vol. 118 (1) , 36
A Three-Stage Transformer-Based Approach for Food Mass Estimation
Accurate food mass estimation is a key component of automated calorie estimation tools, and there is growing interest in leveraging image analysis for this purpose due to its ease of use and scalability. However, current methods face important limitations. Some rely on 3D sensors for depth estimation, which are not widely accessible to all users, while others depend on camera intrinsic parameters to estimate volume, reducing their adaptability across different devices. Furthermore, AI-based approaches that bypass these parameters often struggle with generalizability when applied to images captured using diverse sensors or camera settings. To overcome these challenges, we introduce a three-stage, transformer-based method for estimating food mass from RGB images, balancing accuracy, computational efficiency, and scalability. The first stage applies the Segment Anything Model (SAM 2) to segment food items in images from the SUECFood dataset. Next, we use the Global-Local Path Network (GLPN) to perform monocular depth estimation (MDE) on the Nutrition5k dataset, inferring depth information from a single image. These outputs are then combined through alpha compositing to generate enhanced composite images with precise object boundaries. Finally, a Vision Transformer (ViT) model processes the composite images to estimate food mass by extracting relevant visual and spatial features. Our method achieves notable improvements in accuracy compared to previous approaches, with a mean squared error (MSE) of 5.61 and a mean absolute error (MAE) of 1.07. Notably, this pipeline does not require specialized hardware like depth sensors or multi-view imaging, making it well-suited for practical deployment. Future work will explore the integration of ingredient recognition to support a more comprehensive dietary assessment system.
- 7 November 2025
- Published by MDPI AG in Engineering Proceedings
- Vol. 118 (1) , 36
Open AccessA Three-Stage Transformer-Based Approach for Food Mass Estimation
Accurate food mass estimation is a key component of automated calorie estimation tools, and there is growing interest in leveraging image analysis for this purpose due to its ease of use and scalability. However, current methods face important limitations. Some rely on 3D sensors for depth estimation, which are not widely accessible to all users, while others depend on camera intrinsic parameters to estimate volume, reducing their adaptability across different devices. Furthermore, AI-based approaches that bypass these parameters often struggle with generalizability when applied to images captured using diverse sensors or camera settings. To overcome these challenges, we introduce a three-stage, transformer-based method for estimating food mass from RGB images, balancing accuracy, computational efficiency, and scalability. The first stage applies the Segment Anything Model (SAM 2) to segment food items in images from the SUECFood dataset. Next, we use the Global-Local Path Network (GLPN) to perform monocular depth estimation (MDE) on the Nutrition5k dataset, inferring depth information from a single image. These outputs are then combined through alpha compositing to generate enhanced composite images with precise object boundaries. Finally, a Vision Transformer (ViT) model processes the composite images to estimate food mass by extracting relevant visual and spatial features. Our method achieves notable improvements in accuracy compared to previous approaches, with a mean squared error (MSE) of 5.61 and a mean absolute error (MAE) of 1.07. Notably, this pipeline does not require specialized hardware like depth sensors or multi-view imaging, making it well-suited for practical deployment. Future work will explore the integration of ingredient recognition to support a more comprehensive dietary assessment system.
- Open Access
- 7 November 2025
- Published by MDPI AG in Engineering Proceedings
- Vol. 118 (1) , 46
Evaluating Voice Biomarkers and Deep Learning for Neurodevelopmental Disorder Screening in Real-World Conditions
Voice acoustics have been extensively investigated as potential non-invasive markers for Autism Spectrum Disorder (ASD). Although many studies report high accuracies, they typically rely on highly controlled clinical protocols that reduce linguistic variability. Their data is also recorded using specialized microphone arrays that ensure high-quality recordings. Such dependencies limit their applicability in real-world or in-home screening contexts. In this work, we explore an alternative approach designed to reflect the requirements of mobile-based applications that could assist parents in monitoring their children. We use an open-access dataset of naturalistic storytelling, extracting only the speech segments in which the child is speaking. We applied previously published ASD voice-analysis pipelines to this dataset, which yielded suboptimal performance under these less controlled conditions. We then introduce a deep learning-based method that learns discriminative representations directly from raw audio, eliminating the need for manual feature extraction while being more robust to environmental noise. This approach achieves an accuracy of up to 77% in classifying children with ASD, children with Attention Deficit Hyperactivity Disorder (ADHD), and neurotypical children. Frequency-band occlusion sensitivity analysis on the deep model revealed that ASD speech relied more heavily on the 2000–4000 Hz range, TD speech on both low (100–300 Hz) and high (4000–8000 Hz) bands, and ADHD speech on mid-frequency regions. These spectral patterns may help bring us closer to developing practical, accessible pre-screening tools for parents.
- 7 November 2025
- Published by MDPI AG in Engineering Proceedings
- Vol. 118 (1) , 46
Open AccessEvaluating Voice Biomarkers and Deep Learning for Neurodevelopmental Disorder Screening in Real-World Conditions
Voice acoustics have been extensively investigated as potential non-invasive markers for Autism Spectrum Disorder (ASD). Although many studies report high accuracies, they typically rely on highly controlled clinical protocols that reduce linguistic variability. Their data is also recorded using specialized microphone arrays that ensure high-quality recordings. Such dependencies limit their applicability in real-world or in-home screening contexts. In this work, we explore an alternative approach designed to reflect the requirements of mobile-based applications that could assist parents in monitoring their children. We use an open-access dataset of naturalistic storytelling, extracting only the speech segments in which the child is speaking. We applied previously published ASD voice-analysis pipelines to this dataset, which yielded suboptimal performance under these less controlled conditions. We then introduce a deep learning-based method that learns discriminative representations directly from raw audio, eliminating the need for manual feature extraction while being more robust to environmental noise. This approach achieves an accuracy of up to 77% in classifying children with ASD, children with Attention Deficit Hyperactivity Disorder (ADHD), and neurotypical children. Frequency-band occlusion sensitivity analysis on the deep model revealed that ASD speech relied more heavily on the 2000–4000 Hz range, TD speech on both low (100–300 Hz) and high (4000–8000 Hz) bands, and ADHD speech on mid-frequency regions. These spectral patterns may help bring us closer to developing practical, accessible pre-screening tools for parents.
- Open Access
- 22 October 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (11) , 1136
A Scoping Review of AI-Based Approaches for Detecting Autism Traits Using Voice and Behavioral Data
This scoping review systematically maps the rapidly evolving application of Artificial Intelligence (AI) in Autism Spectrum Disorder (ASD) diagnostics, specifically focusing on computational behavioral phenotyping. Recognizing that observable traits like speech and movement are critical for early, timely intervention, the study synthesizes AI’s use across eight key behavioral modalities. These include voice biomarkers, conversational dynamics, linguistic analysis, movement analysis, activity recognition, facial gestures, visual attention, and multimodal approaches. The review analyzed 158 studies published between 2015 and 2025, revealing that modern Machine Learning and Deep Learning techniques demonstrate highly promising diagnostic performance in controlled environments, with reported accuracies of up to 99%. Despite this significant capability, the review identifies critical challenges that impede clinical implementation and generalizability. These persistent limitations include pervasive issues with dataset heterogeneity, gender bias in samples, and small overall sample sizes. By detailing the current landscape of observable data types, computational methodologies, and available datasets, this work establishes a comprehensive overview of AI’s current strengths and fundamental weaknesses in ASD diagnosis. The article concludes by providing actionable recommendations aimed at guiding future research toward developing diagnostic solutions that are more inclusive, generalizable, and ultimately applicable in clinical settings.
- 22 October 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (11) , 1136
Open AccessA Scoping Review of AI-Based Approaches for Detecting Autism Traits Using Voice and Behavioral Data
This scoping review systematically maps the rapidly evolving application of Artificial Intelligence (AI) in Autism Spectrum Disorder (ASD) diagnostics, specifically focusing on computational behavioral phenotyping. Recognizing that observable traits like speech and movement are critical for early, timely intervention, the study synthesizes AI’s use across eight key behavioral modalities. These include voice biomarkers, conversational dynamics, linguistic analysis, movement analysis, activity recognition, facial gestures, visual attention, and multimodal approaches. The review analyzed 158 studies published between 2015 and 2025, revealing that modern Machine Learning and Deep Learning techniques demonstrate highly promising diagnostic performance in controlled environments, with reported accuracies of up to 99%. Despite this significant capability, the review identifies critical challenges that impede clinical implementation and generalizability. These persistent limitations include pervasive issues with dataset heterogeneity, gender bias in samples, and small overall sample sizes. By detailing the current landscape of observable data types, computational methodologies, and available datasets, this work establishes a comprehensive overview of AI’s current strengths and fundamental weaknesses in ASD diagnosis. The article concludes by providing actionable recommendations aimed at guiding future research toward developing diagnostic solutions that are more inclusive, generalizable, and ultimately applicable in clinical settings.
- 11 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 4561-4568
Multi-Spectral Imaging and Data Fusion for Real-Time Bleeding Detection
In this paper, we introduce a multi-spectral computer vision system that integrates long-wave infrared (thermal) and near-infrared (depth) imaging for real-time bleeding detection. Our methodology leverages deep learning techniques for robust object detection within each imaging modality. Detecting liquids in vision-based systems is particularly challenging due to their irregular shapes and varying intensity patterns. We address these challenges through extensive experimentation. A key aspect of our approach is the development of an optimized annotation strategy that accounts for the thermal properties of liquid patches, improving model training and reducing false positives. Additionally, we analyze how the nature of the surface beneath the liquid, whether soft or solid, influences detection accuracy. To enhance detection consistency, we introduce a mechanism for aligning and matching detected regions across both imaging modalities, which is crucial for distinguishing blood from other warm liquids. Our findings indicate that reducing the number of bounding boxes, by enclosing entire liquid patches rather than segmenting individual droplets, improves detection reliability. Furthermore, we establish an integrated decisionmaking framework that supports data fusion among detected regions of interest across multiple modalities. In a specific use case, this framework is designed to activate an alarm in the event of bleeding. Rigorous experimental validation demonstrates the effectiveness of the proposed system in accurately detecting blood and differentiating it from other liquid spills across various practical applications.
- 11 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 4561-4568
Multi-Spectral Imaging and Data Fusion for Real-Time Bleeding Detection
In this paper, we introduce a multi-spectral computer vision system that integrates long-wave infrared (thermal) and near-infrared (depth) imaging for real-time bleeding detection. Our methodology leverages deep learning techniques for robust object detection within each imaging modality. Detecting liquids in vision-based systems is particularly challenging due to their irregular shapes and varying intensity patterns. We address these challenges through extensive experimentation. A key aspect of our approach is the development of an optimized annotation strategy that accounts for the thermal properties of liquid patches, improving model training and reducing false positives. Additionally, we analyze how the nature of the surface beneath the liquid, whether soft or solid, influences detection accuracy. To enhance detection consistency, we introduce a mechanism for aligning and matching detected regions across both imaging modalities, which is crucial for distinguishing blood from other warm liquids. Our findings indicate that reducing the number of bounding boxes, by enclosing entire liquid patches rather than segmenting individual droplets, improves detection reliability. Furthermore, we establish an integrated decisionmaking framework that supports data fusion among detected regions of interest across multiple modalities. In a specific use case, this framework is designed to activate an alarm in the event of bleeding. Rigorous experimental validation demonstrates the effectiveness of the proposed system in accurately detecting blood and differentiating it from other liquid spills across various practical applications.