Stephanie L. Bennett

In-Bed Mobility Monitoring Using Pressure Sensors

Automated assessment of older adult health is needed due to an impending demographic shift. Mobility is considered an indicator of health and is more tangible than some other health measures. Currently, many papers aim to examine a discrete movement in detail, but none describe one system of algorithms aiming to automatically identify discrete and continuous patient positions and transitions. This paper aims to develop such a system of algorithms. Discrete and continuous data were generated by 32 subjects performing a series of position-transition movements, captured by fiber-optic pressure sensor mats. Algorithm set 1 part 1 aimed to identify and distinguish between three positional states by extracting seven occupancy and dispersion features, then using 1-D and 2-D support vector machine (SVM) and linear classifiers to classify the data. Set 1 part 2 aimed to identify and distinguish between state transitions by calculating percentage pressure difference on a per sensor and large area basis, then monitoring these signals for pressure relief. The second set aimed to examine all movements by extracting six geometric features from center of pressure signals, then using 1-D and 2-D SVM and linear classifiers to classify two subtly different transitions. All methods resulted in at least a 98% identification accuracy, and some methods were able to shed light on the subtleties of transitions. The results suggest that, with more development, the presented algorithmic methods could be implemented in hospital settings to assist with identification and assessment of elderly patient mobility.

Adaptive eulerian video magnification methods to extract heart rate from thermal video

The worlds expanding and aging population has created a demand for inexpensive, unobtrusive, automated healthcare solutions. Eulerian Video Magnification (EVM) aids in the development of these solutions by allowing for the extraction of physiological signals from video data. This paper examines the potential of thermal video in conjunction with EVM to extract physiological measures, particularly heart rate. This paper also proposes an adaptive EVM approach to amplify the desired signal, while avoiding noise amplification. A subject, wearing a textile sensor band collecting ECG, sat still while both a thermal camera and an iPad camera captured video. The iPad video was subjected to EVM, using a wide bandpass filter and low magnification factor. Mean intensity signals for five Regions of Interest (ROIs) were then calculated to extract a signal representing heart rate. The ECG signal was used to validate the ROI resulting in the mean intensity signal best representing heart rate. The thermal video was then subjected to EVM using the same wide bandpass filter and the identified ideal ROI mean intensity post-processing. This signal was compared to the enhanced iPad video mean intensity signal to verify the correct signal was extracted. The original thermal video was subjected again to EVM processing and ROI mean intensity post-processing, this time using an adapted, targeted narrow bandpass filter. Results indicated that thermal video, in conjunction with the proposed adapted EVM method and ROI post-processing can reveal physiological signals like heart rate and limit the potential of revealing an amplified noise signal.

The detection of breathing behavior using Eulerian-enhanced thermal video.

The current gold standard for detecting and distinguishing between types of sleep apnea is expensive and invasive. This paper aims to examine the potential of inexpensive and unobtrusive thermal cameras in the identification and distinction between types of sleep apnea. A thermal camera was used to gather video of a subject performing regular nasal breathing, nasal hyperventilation and an additional trial simulating one type of sleep apnea. Simultaneously, a respiratory inductance plethysmography (RIP) band gathered respiratory data. Thermal video of all three trials were subjected to Eulerian Video Magnification; a procedure developed at MIT for enhancing subtle color variations in video data. Post magnification, nasal regions of interest were defined and mean region intensities were found for each frame of each trial. These signals were compared to determine the best performing region and compared to RIP data to validate breathing behavior. While some regions performed better, all region intensity signals depicted correct breathing behavior. The mean intensity signals for normal breathing and hyperventilation were correct and correlated well with RIP data. Furthermore, the RIP data resulting from the sleep apnea simulation clearly depicted chest movement while the corresponding mean intensity signal depicted lack of cyclical air flow. These results indicate that a subjects breathing behavior can be captured using thermal video and suggest that, with further development and additional equipment, thermal video can be used to detect and distinguish between types of sleep apnea.

Pressure signal feature extraction for the differentiation of clinical mobility assessments

While clinical measures of mobility and balance are important for tracking disease progression in the elderly, most of these tools are based on what can be observed by the human eye, and many do not assess bedridden patients. This paper examines the potential for pressure sensitive mats to be used in conjunction with data processing to develop a system that automates a clinical tool used to assess balance and mobility in the elderly. A study was conducted in which pressure data were gathered while 30 non-patient volunteers performed partial in-bed clinical assessments. Data were then analyzed by grouping sensor data, calculating ratios, then extracting features from the analyzed signals. Pressure ratio signals representing each part of the simulated assessment, were consistent among volunteers and were visually and numerically distinguishable from another. These results indicate that the movement specific pressure signal features identified here are quantifiable and that algorithms may be written to identify and distinguish between certain movements and output the correct clinical assessment.

Distinguishing between stable and unstable sit-to-stand transfers using pressure sensors

The sit-to-stand transfer has been examined in depth using many different methods. Pressure mats and or force plates have been used to partition the transfer into phases. These phases help to identify the beginning and end of the transfer and describe the movement performance, therefore characterizing the transition. The methods of phase breakdown, however, vary depending on the source. In this paper, the objective was to distinguish between a vigorous sit to stand transfer and a slower, less stable sit to stand transfer using a different approach. Ten subjects performed two sit-to-stand transfers each on a hospital mattress, while pressure sensors underneath the mattress gathered movement data. The center of mass was calculated and three features were extracted, then evaluated by a classifier to determine if the features could distinguish between the two movements. Examination of the results determined that two of the three features yielded total accuracy, despite inconsistencies in subject performance. These results suggest that the center of pressure can be used to distinguish between slow, unstable sit-to-stand transfers and vigorous ones. Future work will include the examination of center of pressure in the measurement of balance, as well as the overall assessment of sit-to-stand performance.

Monitoring the relief of pressure points for pressure ulcer prevention: A subject dependent approach

Pressure ulcers are of great cost to both the patient and the healthcare system. Devices have been developed with the goal of pressure ulcer prevention, but many available technically complex devices have been shown to be no more effective than low pressure overlays or mattresses. This paper proposes a subject dependent algorithm capable of automatically detecting when and where pressure points have been relieved from underneath a supine subject, without any user inputs or assumptions. Pressure sensitive mats, associated software, a laptop and a video camera were used to measure and collect pressure signals generated by a supine subject performing 3 movements: the subject rolling to one side of the body, then to the other side, and the subject attempting to roll without lifting any pressure points off the mattress. The data was zeroed, baseline values were found, differences in sensor score from baseline were calculated, and instances during which a valley on one side coincided with a peak on the other, were recorded. Examination of these results indicated that the algorithm was capable of determining when and where pressure points underneath the sacrum and foot regions were lifted off the bed, but not capable of determining if a scapula pressure point was relieved. These results suggest that the proposed algorithm is effective for some, but not all regions of the body. Future work will therefore focus on detection of all pressure points, and the adjustment of the algorithm for subject independence.

The use of a thermal camera and Eulerian enhancement in the examination of pedal pulse and microvascular health

Early detection of impaired blood flow and microvascular functioning is important to prevent ulceration in diabetic patients. This paper aims to first determine if thermal video in conjunction with Eulerian Video Magnification (EVM) can be used to find the pedal pulse rate, and reveal patterns indicative of the foots microvascular health. Thermal video was captured of a healthy adults foot while a Doppler ultrasound captured pedal pulse. Another thermal video was captured of a patients heels. These videos were subjected to EVM, areas of interest were defined and the mean intensity signal was calculated temporally, within each defined area. The healthy adult signals were compared to Doppler data to determine the signal best representative of pedal pulse. The patient signals were examined for patterns. The mean intensity signals best representing pedal pulse in the healthy adult resulted from areas containing an artery close to the skin. The most significant pattern in the patient data was a large difference in signal amplitude from areas containing the left posterior tibial artery and the right; the left, colder heel had a weaker signal amplitude. These results suggest that thermal video subjected to EVM can reveal the pedal pulse rate by extracting intensity signals from areas in which arteries are close to the skin, and may reveal differences in the microvascular health of the left versus right foot. The ability to detect pedal pulse and differences in microvascular health using an inexpensive and non-intrusive thermal camera would of great value to a podiatric clinic.Early detection of impaired blood flow and microvascular functioning is important to prevent ulceration in diabetic patients. This paper aims to first determine if thermal video in conjunction with Eulerian Video Magnification (EVM) can be used to find the pedal pulse rate, and reveal patterns indicative of the foots microvascular health. Thermal video was captured of a healthy adults foot while a Doppler ultrasound captured pedal pulse. Another thermal video was captured of a patients heels. These videos were subjected to EVM, areas of interest were defined and the mean intensity signal was calculated temporally, within each defined area. The healthy adult signals were compared to Doppler data to determine the signal best representative of pedal pulse. The patient signals were examined for patterns. The mean intensity signals best representing pedal pulse in the healthy adult resulted from areas containing an artery close to the skin. The most significant pattern in the patient data was a large difference in signal amplitude from areas containing the left posterior tibial artery and the right; the left, colder heel had a weaker signal amplitude. These results suggest that thermal video subjected to EVM can reveal the pedal pulse rate by extracting intensity signals from areas in which arteries are close to the skin, and may reveal differences in the microvascular health of the left versus right foot. The ability to detect pedal pulse and differences in microvascular health using an inexpensive and non-intrusive thermal camera would of great value to a podiatric clinic.

Measurements of change in thermal images due to applied pressure

Thermal imaging is of value to medical professionals because of its low risk and non-invasive properties. While thermal imaging has been explored in the area of pressure ulcers, many relevant papers address existing pressure ulcers and few address the prevention of pressure ulcers. This paper aims to examine the potential of thermal imaging in the prevention of pressure ulcers by extracting temperature-based and region-based measurements from thermal images and quantifying thermal patterns. A subject was asked to press on a pressure sensor mat at two specified intensities, and a series of thermal images were taken before and after to track thermal behaviour. These images were subjected to standard image processing techniques before temperature specific contour and area measurements were extracted as well as region specific intensity and weighted centroid measurements. Results indicated that the contour and area measurements were able to capture the temperature pattern of the whole hand, while the intensity measurements were able to indicate region specific thermal patterns. These results suggest that the extraction of measurements from a series of thermal images can capture and quantify visually identifiable thermal patterns of the hand over time. These findings will be expanded upon in future work by further examining different measurements, sharper images, different equipment and the involvement of elderly patients. While future collection of patient data is expected to yield different thermal patterns, this paper has demonstrated recognition and quantification of a pattern, regardless of the pattern itself.

Comparison of motion-based analysis to thermal-based analysis of thermal video in the extraction of respiration patterns

Non-contact methods of extracting vital signals has become a popular area of research. This is likely due to the worlds aging population and the increased need for long term and remote monitoring. This paper examines and compares the potential for one modality to capture a vital sign, specifically respiration, in the presence of signal abnormalities. This paper compares temperature based-methods to motion-based methods of extracting respiration rate from thermal video of a subject performing computationally difficult respiration tests. The thermal video was subjected to segmentation-based image processing and region tracking to encompass temperature changes over time. All methods were successful in identifying regular breathing and the absence of breathing, but differed in performance identifying hyperventilation and obstructive sleep apnea simulated breathing. The temperature-based method better depicted airflow volume, while the motion-based method better depicted absence of breath and chest movement; neither signal on its own was able to accurately depict OSA breathing. These results suggest that the fusion of information from different physical phenomenon (i.e. motion and temperature) is important here in detecting abnormal breathing patterns, but also in the detection of all vital signals, adding algorithmic robustness in the presence of signal abnormalities.

Adaptive Eulerian Video Processing of Thermal Video: An Experimental Analysis

The use of spatiotemporal video processing to extract biosignals is an emerging technique. This paper aims to build upon current work through robust experimentation and analysis. A blood flow simulation model was captured by thermal and optical cameras, while hot water was pumped through the system. Additionally, five subjects were recruited to perform two experimental trials: a facial perfusion trial and an arm blood occlusion trial, for which subjects sat quietly, while video data were captured using thermal and optical cameras. Each video was subjected to region of interest selection and adaptive Eulerian video magnification (EVM); the iterative application of EVM, first with a wide temporal bandpass filter and low amplification factor and again with a narrower, targeted temporal bandpass filter and higher amplification factor. The results from the simulation experiments indicated that thermal video in conjunction with adaptive EVM processing can reveal variations in temperature indicative of pulse rate in a controlled system of known variables. This process helped to better characterize Eulerian signal enhancement versus Eulerian noise enhancement. The results from the facial perfusion experiments suggest that the adaptive EVM processing of thermal video results in signals representative of facial perfusion rate. The results from the blood occlusion experiments revealed an occlusion temperature pattern, but not a perfusion rate. This paper therefore further demonstrated the potential of thermal video in conjunction with adaptive EVM methods to extract a signal representative of facial perfusion rate, and illustrated the need for more research on thermal video and adaptive EVM.