Yasmina Souley Dosso
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yasmina Souley Dosso has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- James Robert Green
Department of Systems and Computer Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada
5 shared publications - Kim Greenwood
Department of Mechanical Engineering, University of Ottawa, Ottawa, Canada
5 shared publications - JoAnn Harrold
Neonatology, Children’s Hospital of Eastern Ontario, Ottawa, ON K1H 8L1, Canada
4 shared publications Amente Bekele
Systems and Computer Engineering, Carleton University, Ottawa, Canada
1 shared publicationCheryl Aubertin
Neonatology, Children’s Hospital of Eastern Ontario, Ottawa, Canada
1 shared publication- Daniel G. Kyrollos
Department of Systems and Computer Engineering, Carleton University Biomedical Informatics Collaboratory, Carleton University, Ottawa, ON, Canada
1 shared publication
Latest Publications
- Open Access
Depth-Based Intervention Detection in the Neonatal Intensive Care Unit Using Vision Transformers
Depth cameras can provide an effective, noncontact, and privacy-preserving means to monitor patients in the Neonatal Intensive Care Unit (NICU). Clinical interventions and routine care events can disrupt video-based patient monitoring. Automatically detecting these periods can decrease the time required for hand-annotating recordings, which is needed for system development. Moreover, the automatic detection can be used in the future for real-time or retrospective intervention event classification. An intervention detection method based solely on depth data was developed using a vision transformer (ViT) model utilizing real-world data from patients in the NICU. Multiple design parameters were investigated, including encoding of depth data and perspective transform to account for nonoptimal camera placement. The best-performing model utilized ∼85 M trainable parameters, leveraged both perspective transform and HHA (Horizontal disparity, Height above ground, and Angle with gravity) encoding, and achieved a sensitivity of 85.6%, a precision of 89.8%, and an F1-Score of 87.6%.
Open AccessDepth-Based Intervention Detection in the Neonatal Intensive Care Unit Using Vision Transformers
Depth cameras can provide an effective, noncontact, and privacy-preserving means to monitor patients in the Neonatal Intensive Care Unit (NICU). Clinical interventions and routine care events can disrupt video-based patient monitoring. Automatically detecting these periods can decrease the time required for hand-annotating recordings, which is needed for system development. Moreover, the automatic detection can be used in the future for real-time or retrospective intervention event classification. An intervention detection method based solely on depth data was developed using a vision transformer (ViT) model utilizing real-world data from patients in the NICU. Multiple design parameters were investigated, including encoding of depth data and perspective transform to account for nonoptimal camera placement. The best-performing model utilized ∼85 M trainable parameters, leveraged both perspective transform and HHA (Horizontal disparity, Height above ground, and Angle with gravity) encoding, and achieved a sensitivity of 85.6%, a precision of 89.8%, and an F1-Score of 87.6%.
- Open Access
- 18 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 144168-144179
Uncertainty Measurements in Non-Contact Neonatal Heart Rate Monitoring
Continuously monitoring patient vital signs in the neonatal intensive care unit (NICU) requires wired sensors that can irritate fragile skin, motivating the development of non-contact physiologic signal estimation approaches. Such estimators typically involve a pipeline of multiple data analysis stages, including pre-processing, region of interest detection and tracking, and physiologic parameter estimation. Uncertainty in the estimated physiologic signal is often quantified strictly from the signal quality indicator (SQI) generated by the final stage of the pipeline. This manuscript proposes a framework to account for SQIs generated by each stage in a physiologic signal estimation pipeline and using the fusion of all SQIs to arrive at a refined measure of uncertainty in the final estimated physiologic parameter. This framework is demonstrated for heart rate (HR) estimation of newborns admitted in the NICU, where the pipeline includes bed occupancy detection, face detection, face tracking, and physiologic signal estimation. Different SQIs are derived at each stage and a novel fusion of all SQI metrics is shown to produce a more effective estimate of uncertainty in the final estimated HR.
- 18 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 144168-144179
Open AccessUncertainty Measurements in Non-Contact Neonatal Heart Rate Monitoring
Continuously monitoring patient vital signs in the neonatal intensive care unit (NICU) requires wired sensors that can irritate fragile skin, motivating the development of non-contact physiologic signal estimation approaches. Such estimators typically involve a pipeline of multiple data analysis stages, including pre-processing, region of interest detection and tracking, and physiologic parameter estimation. Uncertainty in the estimated physiologic signal is often quantified strictly from the signal quality indicator (SQI) generated by the final stage of the pipeline. This manuscript proposes a framework to account for SQIs generated by each stage in a physiologic signal estimation pipeline and using the fusion of all SQIs to arrive at a refined measure of uncertainty in the final estimated physiologic parameter. This framework is demonstrated for heart rate (HR) estimation of newborns admitted in the NICU, where the pipeline includes bed occupancy detection, face detection, face tracking, and physiologic signal estimation. Different SQIs are derived at each stage and a novel fusion of all SQI metrics is shown to produce a more effective estimate of uncertainty in the final estimated HR.
- Open Access
- 11 July 2022
- Published by Carleton University
Machine Vision for Patient Monitoring in the Neonatal Intensive Care Unit
Continuous patient monitoring of newborns in the neonatal intensive care unit (NICU) is often performed with wired sensors which can be cumbersome, can interfere with parental bonding, and can irritate the patient's fragile skin. Non-contact video-based patient monitoring systems are therefore a preferrable solution. While a multitude of high-performing machine vision technologies have been successfully implemented on an adult population, such methods often fail in neonatal population. In this thesis, we assess state-of-the-art adult-based methods to bridge the gap to an understudied neonatal population in the NICU environment. To this end, several important machine vision concepts are investigated, including scene understanding, image classification, face detection, semantic segmentation, motion detection, face tracking, and heart rate estimation. In each of these areas, we assess the state-of-the-art and identify its applicability to a neonatal population. In cases where serious limitations are observed, this thesis pushes the state-of-the-art and implements new techniques more suitable for newborns. Finally, a non-contact neonatal heart rate monitoring pipeline is created using multiple research contributions in this thesis. Doing so, we obtain a vital sign decision support tool for clinical use by estimating the uncertainty in each research contributions and demonstrating how errors can propagate from one to another. Thirty-three newborns admitted to the NICU at the Children's Hospital of Eastern Ontario were recorded using a depth-sensing camera, which simultaneously captures color, depth and near-infrared videos, thereby acquiring pertinent data in all lighting conditions. Gold standard event annotations and physiologic data were recorded simultaneously as ground truth data for the development of machine vision models. Our proposed approach includes a combination of machine vision, deep learning, image processing, and signal processing techniques to overcome environmental factors such as lighting variations, occlusion, and motion artifacts. This thesis implements an efficient, robust, and reliable prototype neonatal monitoring system for potential future deployment in hospital settings. To this end, this research aimed to exploit transfer learning from state-of-the-art models to address problems such as complex NICU scenes, variations in newborn's visual features, data scarcity, and class imbalance which are often observed in clinical research and neonatal monitoring applications.
- 11 July 2022
- Published by Carleton University
Open AccessMachine Vision for Patient Monitoring in the Neonatal Intensive Care Unit
Continuous patient monitoring of newborns in the neonatal intensive care unit (NICU) is often performed with wired sensors which can be cumbersome, can interfere with parental bonding, and can irritate the patient's fragile skin. Non-contact video-based patient monitoring systems are therefore a preferrable solution. While a multitude of high-performing machine vision technologies have been successfully implemented on an adult population, such methods often fail in neonatal population. In this thesis, we assess state-of-the-art adult-based methods to bridge the gap to an understudied neonatal population in the NICU environment. To this end, several important machine vision concepts are investigated, including scene understanding, image classification, face detection, semantic segmentation, motion detection, face tracking, and heart rate estimation. In each of these areas, we assess the state-of-the-art and identify its applicability to a neonatal population. In cases where serious limitations are observed, this thesis pushes the state-of-the-art and implements new techniques more suitable for newborns. Finally, a non-contact neonatal heart rate monitoring pipeline is created using multiple research contributions in this thesis. Doing so, we obtain a vital sign decision support tool for clinical use by estimating the uncertainty in each research contributions and demonstrating how errors can propagate from one to another. Thirty-three newborns admitted to the NICU at the Children's Hospital of Eastern Ontario were recorded using a depth-sensing camera, which simultaneously captures color, depth and near-infrared videos, thereby acquiring pertinent data in all lighting conditions. Gold standard event annotations and physiologic data were recorded simultaneously as ground truth data for the development of machine vision models. Our proposed approach includes a combination of machine vision, deep learning, image processing, and signal processing techniques to overcome environmental factors such as lighting variations, occlusion, and motion artifacts. This thesis implements an efficient, robust, and reliable prototype neonatal monitoring system for potential future deployment in hospital settings. To this end, this research aimed to exploit transfer learning from state-of-the-art models to address problems such as complex NICU scenes, variations in newborn's visual features, data scarcity, and class imbalance which are often observed in clinical research and neonatal monitoring applications.
- Open Access
- 8 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 62893-62909
NICUface: Robust Neonatal Face Detection in Complex NICU Scenes
The development of non-contact patient monitoring applications for the neonatal intensive care unit (NICU) is an active research area, particularly in facial video analysis. Recent studies have used facial video data to estimate vital signs, assess pain from facial expression, differentiate sleep-wake status, detect jaundice, and in face recognition. These applications depend on an accurate definition of the patient’s face as a region of interest (ROI). Most studies have required manual ROI definition, while others have leveraged automated face detectors developed for adult patients, without systematic validation for the neonatal population. To overcome these issues, this paper first evaluates the state-of-the-art in face detection in the NICU setting. Finding that such methods often fail in complex NICU environments, we demonstrate how fine-tuning can increase neonatal face detector robustness, resulting in our NICUface models. A large and diverse neonatal dataset was gathered from actual patients admitted to the NICU across three studies and gold standard face annotations were completed. In comparison to state-of-the-art face detectors, our NICUface models address NICU-specific challenges such as ongoing clinical intervention, phototherapy lighting, occlusions from hospital equipment, etc. These analyses culminate in the creation of robust NICUface detectors with improvements on our most challenging neonatal dataset of +36.14, +35.86, and +32.19 in AP30, AP50, and mAP respectively, relative to state-of-the-art CE-CLM, MTCNN, img2pose, RetinaFace, and YOLO5Face models. Face orientation estimation is also addressed, leading to an accuracy of 99.45%. Fine-tuned NICUface models, gold-standard face annotation data, and the face orientation estimation method are also released here.
- 8 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 62893-62909
Open AccessNICUface: Robust Neonatal Face Detection in Complex NICU Scenes
The development of non-contact patient monitoring applications for the neonatal intensive care unit (NICU) is an active research area, particularly in facial video analysis. Recent studies have used facial video data to estimate vital signs, assess pain from facial expression, differentiate sleep-wake status, detect jaundice, and in face recognition. These applications depend on an accurate definition of the patient’s face as a region of interest (ROI). Most studies have required manual ROI definition, while others have leveraged automated face detectors developed for adult patients, without systematic validation for the neonatal population. To overcome these issues, this paper first evaluates the state-of-the-art in face detection in the NICU setting. Finding that such methods often fail in complex NICU environments, we demonstrate how fine-tuning can increase neonatal face detector robustness, resulting in our NICUface models. A large and diverse neonatal dataset was gathered from actual patients admitted to the NICU across three studies and gold standard face annotations were completed. In comparison to state-of-the-art face detectors, our NICUface models address NICU-specific challenges such as ongoing clinical intervention, phototherapy lighting, occlusions from hospital equipment, etc. These analyses culminate in the creation of robust NICUface detectors with improvements on our most challenging neonatal dataset of +36.14, +35.86, and +32.19 in AP30, AP50, and mAP respectively, relative to state-of-the-art CE-CLM, MTCNN, img2pose, RetinaFace, and YOLO5Face models. Face orientation estimation is also addressed, leading to an accuracy of 99.45%. Fine-tuned NICUface models, gold-standard face annotation data, and the face orientation estimation method are also released here.
- Open Access
- 1 November 2021
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 138, 104873
RGB-D scene analysis in the NICU
- 1 November 2021
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 138, 104873
Open AccessRGB-D scene analysis in the NICU
- 1 June 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Real-time Neonatal Respiratory Rate Estimation using a Pressure-Sensitive Mat
We present an approach for real-time respiratory rate (RR) estimation in a neonatal intensive care unit (NICU) using a pressure-sensitive mat (PSM). Real patient data were collected in an NICU from four sources simultaneously: a PSM placed under the patient, a Draeger patient monitor, a video camera placed directly above the patient, and a custom bedside event annotation application running on a tablet. The PSM data were used to develop an algorithm for estimating the patient's RR. The results were evaluated against impedance pneumography (IP) based RR measurements from the patient monitor. In comparison to the IP estimates, we achieved a mean absolute error of 4.51 breaths per minute (bpm) for 3 hours of data collected from a single patient. Moreover, we show that our newer approach performs better than the previous frequency-based approach that was developed based on patient simulator data. The estimation accuracy achieved by the new algorithm meets clinical requirements for the determination of RR.
- 1 June 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Real-time Neonatal Respiratory Rate Estimation using a Pressure-Sensitive Mat
We present an approach for real-time respiratory rate (RR) estimation in a neonatal intensive care unit (NICU) using a pressure-sensitive mat (PSM). Real patient data were collected in an NICU from four sources simultaneously: a PSM placed under the patient, a Draeger patient monitor, a video camera placed directly above the patient, and a custom bedside event annotation application running on a tablet. The PSM data were used to develop an algorithm for estimating the patient's RR. The results were evaluated against impedance pneumography (IP) based RR measurements from the patient monitor. In comparison to the IP estimates, we achieved a mean absolute error of 4.51 breaths per minute (bpm) for 3 hours of data collected from a single patient. Moreover, we show that our newer approach performs better than the previous frequency-based approach that was developed based on patient simulator data. The estimation accuracy achieved by the new algorithm meets clinical requirements for the determination of RR.