Tracking sugar intake has become a trending practice, much like keeping track of calorie intake. However, consumer-grade systems or devices for measuring or detecting sugar levels of unlabeled food items are close to none. In this paper, we discuss how low-cost Multi-Spectral Imaging (MSI) technology can be used to identify the sugar level of grapes in a simple, yet effective manner. Here, grapes are used as a representative for fruit or food. Freshly harvested grapes were used for spectral data collection. Unispectral portable multispectral camera EVK – UNS52000 which has a spectral range of 700-940 nm across 10 spectral bands was used for data collection. Actual sugar content was determined using a handheld refractometer and used for reference. We evaluated the performance of two well-known machine learning-based regression algorithms such as Random Forest (RF) and Light Gradient Boosting Machines (LightGBM). Spectral data derived from images captured in 10 bands were used as features or independent variables and actual sugar measured using a refractometer was used as a reference or the dependent variable in regression models. We tested the model performance using different feature scenarios: a) All 10 spectral bands, b) Top 5 spectral bands and c) Top 3 spectral bands. Results showed that R2 was 0.92 for both RF and LightGBM with negligible variation in RMSE (1.05-1.06 °Brix) when using all 10 bands. There was minimal difference in terms of R2 and RMSE. However, there was a significant difference in model training time with LightGBM being much faster than RF. Moreover, by selecting the top 5 and 3 features, we reduced the R2 to 0.89 and 0.78 respectively in the case of LightGBM. While there is a reduction in R2, selecting the top 5 bands will be useful from an operational perspective in terms of computational power and cost of device development. We aim to design and develop the device using the selected top 5 bands for operational on-the-fly prediction of sugar content in grapes. The proposed approach could be helpful for various stakeholders such as people with Hypoglycemia or Diabetes, grape producers to gauge the maturity of grapes, and industrial use cases for vineyards.

Post-pandemic, wellness and healthcare sector is vocal about balanced food consumption, inclusion of fresh fruits and regular exercise. Freshness and sugar contents of fruits are crucial to understand before their consumption. A normal RGB camera can provide information about the freshness of fruits based on the surface composition of images generated by the camera. Interestingly, the Multi-Spectral Imaging (MSI) provides information that is superior to a standard RGB camera, as it considers the NIR band. In this article, the authors have deliberated how MSI can be used to predict the sugar level and freshness in fruits, particularly grapes using modified EfficientNet-B0 as a Deep Learning (DL) model for the analysis. For predicting sugar level, the average value across the 5-fold cross-validation (cv) achieved by the DL model was RMSE of 2.68 (std 0.33) ◦Bx, MAE of 2.11 (0.34) ◦Bx, MAPE of 12% (2%) using MSI and RMSE of 7.76 (1.79) ◦Bx, MAE of 6.65 (1.69) ◦Bx, MAPE of 41% (9%) using RGB camera images. For freshness prediction, the DL model achieved an average 5-fold cv accuracy of 88.35% (9.71%) using MSI and 82.22% (7.9%) using RGB camera images. Results indicated that MSI can predict both the sugar level and freshness whereas RGB camera images can be used only for predicting the freshness but not the sugar level prediction of the grapes. The findings suggest that MSI offers a valuable and versatile solution for quality assessment in the fruit industry, enabling better dietary choices and healthcare regimes.

Stress management is paramount to maintaining optimal health and well-being; stress builds up in spikes, causing problems like hypertension and anxiety, necessitating personalized interventions delivered in real-time through wearable technology. This work underscores the pivotal role of unobtrusive stress detection and presents the development of auto-generated compact models tailored for on-device inference through Neural Architecture Search (NAS). These models aim to facilitate efficient stress monitoring directly on low-power devices, representing a promising avenue for advancing personalized healthcare with continuous monitoring and effective digital interventions.

Patients surviving a brain stroke often experience an impaired neuro-motor coordination and cognitive capability. Rehabilitation assisted by therapists can help them regain much of this coordination. In recent past, gamified activities have proved to be effective in this regard with enhanced patient engagement. But, they may require special hardware and certain usage restrictions regarding patient's positioning etc. for operating effectively. We have set out to develop a suite of games using a smartwatch to sense arm's motion to eliminate such challenges, supporting customized scoring for different post-stroke mobility stages and focusing on state-of-the-art 3D graphics offering a high realism aimed at better patient engagement and adherence. In this demonstration, we present two of these games. First is a 'Forest Stroll' game where the patient controls heading of an in-game character moving on a forest path, with scoring based on how closely the path is followed, with a few additional arm gestures which can be activated by the therapist optionally. Second is a 'Drumming' game, where the patient controls a virtual drumstick and tries to play a customizable pre-saved pattern involving multiple ranges of motion, scored on how closely the pattern is played, testing both motor and cognitive capabilities.

Freezing of Gait (FoG) is one of the incapacitating motor symptoms that appear in patients with Parkinson’s Disease (PD). FoG manifests gait impairments and imposes unforeseen difficulties in commencing the locomotion. Frequent episodes of FoG often lead to fall-related injuries and impart dreadful health repercussions. Prediction of FoG before the occurrence could potentially increase the opportunities for pre-emptive cueing to mitigate or abate the episode of FoG. This paper presents a novel algorithm to predict the onset of FoG using a single ankle accelerometer sensor. Principally, we have focused on designing a lightweight algorithm to facilitate the real-time prediction of FoG in resource-limited hardware. A novel Genetic Algorithm (GA) is introduced as a feature selection method to enhance the algorithm’s performance. We have adapted the patient-invariant model which is a more viable approach in practical deployment. The algorithm is evaluated using the Daphnet dataset and achieved 88% FoG prediction accuracy with 1 second prediction time.

Embedded Artificial Intelligence (AI) is becoming increasingly important in the field of healthcare where such AI enabled devices are utilized to assist physicians, clinicians, and surgeons in their diagnosis, rehabilitation and therapy planning. However, it is still a challenging task to come up with an accurate and efficient machine learning model for resource-limited devices that work$24\times 7$.. It requires both intuition and experience. This dependence on human expertise and reliance on trial-and-error-based design methods create impediments to the standard processes of effort estimation, design phase planning, and generating service-level agreements for projects that involve AI-enabled MedTech devices. In this paper, we present AutoML search from an algorithmic perspective, instead of a more prevalent optimization or black-box tool view. We briefly present and point to case studies that demonstrate the efficacy of the automation approach in terms of productivity improvements. We believe that our proposed method can make AutoML more amenable to the applications of software engineering principles and also accelerate biomedical device engineering, where there is a high dependence on skilled human resources.

Customized, on-device applications that provide timely interventions about smoking episodes are very helpful for smoking cessation. For this, real-time detection of smoking puffs are necessary through unobtrusive wearable devices. This work demonstrates auto-generated tiny puff detection models for on-device inference on low-power wearable devices.

Atrial Fibrillation (AF) is a common cardiac arrhythmia with significant health implications. To evaluate the severity of AF and monitor the effectiveness of treatments, continuous measurement of AF Burden (AFB) is crucial. This demo showcases a cost-effective, portable, and user-friendly solution for AFB monitoring. Our system integrates the Arduino Nano 33 BLE sense board with the AD8232 sensor to reliably capture single-lead ECG data. Additionally, a custom algorithm computes real-time AFB, providing insights into the frequency and duration of AF episodes. This system has the potential to enhance AF management and patient care, making cardiac health monitoring more accessible.

In this work, we demonstrate a radically novel approach towards inertial-only tracking of wrist in real-time on a smartwatch for air-writing tasks. Deriving motion trajectories from commercial-grade Inertial Measurement Units (IMU) has always been a challenging task due to inherent sensor errors and associated trajectory drift. Computationally expensive solutions offered in literature cannot be used for a fully real-time tracking while also maintaining acceptable accuracy. This work presents 'NNTrak', marking our attempt to address these issues using a Convolutional Neural Network (CNN), which is trained to learn various strokes of the wrist and efficiently generates motion trajectory in real-time for air-writing. For this demonstration, we show computationally constrained Raspberry Pi 3B running our solution and a smartwatch worn while drawing a gesture in air with the trajectory being displayed in true real-time.

Continuous monitoring of cardiac health through single-lead wearable Electrocardiogram (ECG), is important for paroxysmal Atrial Fibrillation (AF) detection. Wearable ECG straps, watches, and implantable loop recorders (ILR) are based on this paradigm. These devices are used by medical professionals to view data from multiple patients, perform continuous monitoring and analysis to provide immediate care to patients. These monitoring devices display simple health screening alerts to the subjects and generate distress signals for people working outdoors or in isolated environments with intermittent Internet connectivity. Hence, low-memory, low-power, low-latency on-device inference becomes very important. This work aims at realizing such solutions by providing a framework to generate tiny (less than 256 KB) Deep Neural Networks customized for typical microcontrollers (MCU) used in those devices.

This paper presents a novel approach of generating synthetic Photoplethysmogram (PPG) data using a physical model of the cardiovascular system to improve classifier performance with a combination of synthetic and real data. The physical model is an in-silico cardiac computational model, consisting of a four-chambered heart with electrophysiology, hemodynamic, and blood pressure auto-regulation functionality. Starting with a small number of measured PPG data, the cardiac model is used to synthesize healthy as well as PPG time-series pertaining to coronary artery disease (CAD) by varying pathophysiological parameters. A Variational Autoencoder (VAE) structure is proposed to derive a statistical feature space for CAD classification. Results are presented in two perspectives namely, (i) using artificially reduced real disease data and (ii) using all the real disease data. In both cases, by augmenting with the synthetic data for training, the performance (sensitivity, specificity) of the classifier changes from (i) (0.65, 1) to (1, 0.9) and (ii) (1, 0.95) to (1, 1). The proposed hybrid approach of combining physical modelling and statistical feature space selection generates realistic PPG data with pathophysiological interpretation and can outperform a baseline Generative Adversarial Network (GAN) architecture with a relatively small amount of real data for training. This proposed method could aid as a substitution technique for handling the problem of bulk data required for training machine learning algorithms for cardiac health-care applications.

Freezing of Gait (FoG) is a paroxysmal and devitalizing symptom associated with Parkinson’s disease (PD). Episodes of FoG impedes gait and augments fall propensity, often leading to serious fall-injury. In this paper, we present a method for online detection of FoG using a wearable motion sensor. The novelty lies in utilizing the Empirical Wavelet Transform for signal denoising and incorporating the two new features to ameliorate the accuracy of the algorithm. Fundamentally, we have focused on a patient-independent model and leveraged a single ankle sensor which makes it a more feasible approach in terms of usability. Our model is evaluated on Daphnet dataset and achieved the average Sensitivity of.95 and Specificity of.70 with only a single sensor, demonstrating its immense potential.

Range Of Motion (ROM) of joints is a key biomarker in assessing the osteokinematics of muskuloskeletal system. Goniometry of ROM is effected by orthopaedists through manual measurement using medical grade goniometers. Such measurement requires personal presence of a medical expert especially in passive goniometry. For a digital estimation of ROM of shoulder and elbow joints, smart wearables embedded with inertial sensors have been used. Although they cover a wide variety of ROM measurements, but fail for measurements made in certain planes, and require sensor specific calibration. Through this work, we aim to demonstrate a calibration-free solution for ROM estimation of shoulder and elbow joints, 'ROMeasure' which can work in any random plane of measurement with a high accuracy even at extremely slow speed of rotation greatly enhancing its practicality in a medical scenario. The demonstration includes a user wearing a smartwatch, and rotating elbow/shoulder joints. Graphs for real-time angle and rotation speed are displayed on a computer screen in real time and at the end of session, final range of motion is calculated. We believe that such a setup can be extrememly useful in a tele-health scenario, and owing to the pervasiveness of smart devices today, it can prove to be a highly convenient yet accurate solution for self-assessment. Our system has been observed to incur MAE of less than 5 degrees in meticulous experimentation performed on different subjects in multiple planes of rotation, even at a rotational speed of under 10 degrees per second.

Abnormal heart sounds may have diverse frequency characteristics depending upon underlying pathological conditions. Designing a binary classifier for predicting normal and abnormal heart sounds using supervised learning requires a lot of training data, covering different types of cardiac abnormalities. In this paper, we propose a semi-supervised approach to solve the problem. A convolutional Variational Autoencoder (VAE) structure is defined for learning the probability distribution of the spectrogram properties of normal heart sounds. The Kullback-Leibler (KL) divergence between the known prior distribution of the VAE and the encoded distribution is taken as an anomaly score for detecting abnormal heart sounds. The proposed approach is evaluated on open access and in-house datasets of Phonocardiogram (PCG) signals, recorded from normal subjects and patients, having cardiovascular diseases, cardiac murmurs and extra heart sounds. Results show that an improved classification performance is achieved in comparison to the existing approaches.

Synthetic data generation has recently emerged as a substitution technique for handling the problem of bulk data needed in training machine learning algorithms. Healthcare, primarily cardiovascular domain is a major area where synthetic physiological data can be used improve accuracy of machine learning algorithm. This paper presents a novel approach of generating synthetic Photoplethysmogram (PPG) data using statistical explosion. Synthetic data is subsequently used to classify Coronary Artery Disease (CAD) using a two stage cascaded classifier. Proposed classifier along with synthetic data removes class bias and provides better accuracy compared to state of art. The proposed data generation and cascaded classifier is generic enough to be used to improve machine learning algorithm on any time series signal.

Coronary Artery Disease (CAD) causes significant global mortality. The recent development in artificial intelligence shows the feasibility of early non-invasive screening of several life- threatening cardiovascular diseases. However, such approaches have been less prolific in diagnosis of CAD due to lack of clinically known definite bio-marker. In this paper, we propose a novel neural network architecture that effectively combines two non-specific CAD markers, 1) anomalous morphology of Electrocardiogram (ECG) waveform and 2) abnormal Heart Rate Variability (HRV). A Convolutional Neural Network (CNN) structure is defined for extraction of morphological ECG features. Another composite structure is defined based on Long Short-Term Memory (LSTM) and a set of hand crafted statistical features for measuring the extent of HRV. The two independent bio-markers are subsequently combined in a hybrid CNN-LSTM architecture for classification of CAD. The proposed approach is evaluated on two datasets, a corpus, selected from the MIMIC II waveform dataset and a partially noisy in-house dataset, recorded using a low-cost ECG sensor. Results show that overall classification accuracy of 93% and 88% are achieved on the two datasets, which outperform the existing approaches.

In this work, we present a solution for an accurate and real-time monitoring of breathing rate (BR) from photoplethysmogram (PPG) signal on both smartphone and smartwatch. Respiration induces multiple modulations in a PPG signal which are difficult to extract from low-quality PPG signal collected using consumer devices. We present an effective method of validating the breathing signal data which is evaluated and compared on an open dataset. The solution is also implemented as a smartphone and smartwatch app to provide an on-device real-time BR, and evaluated on multiple subjects. For the demonstration, we shall show breathing rate and breathing pattern both on smartphone and smartwatch, which can be visualized in real-time on a dashboard.

In this paper, we present "ThermoTrak", a smartphone accessory based, real-time and accurate temperature measurement mechanism, which can be used to screen for fever, which is a manifestation of infectious diseases including the symptoms caused by SARS-CoV-2. Our system accurately identifies face and forehead region from a safe distance of one meter, calculates accurate temperature of forehead with accuracy of ±0.5° C on a linear scale. An AI based algorithm is employed for the purpose of accurately detecting the Region of interest (ROI) (Face & point near center of Forehead) and calculate the absolute temperature within 300 milliseconds.

Wearable health monitoring has become a very familiar term in today'sworld. One of the most popular means ofwearable sensing is photoplethysmogram (PPG). Due to its unobtrusive and ubiquitous nature, it is gaining popularity among people everywhere. Due to the ease of use, the utility of such technology is increasing day by day. However, in theworld of researchers, the accurate estimation of heart rate (HR) in presence of motion artefacts remains an unsolved problem due to the susceptibility of PPG signals to corruption by motion artefacts. The way in which a person fastens the device on the wrist plays an important role in the acquisition of signal from the device. While there are various research works going on in this field, there is always a trade-off between accuracy and complexity of algorithm and hardware resources. Also, in such scenarios where the sensor gets misplaced due to movements, there might be no PPG signal component available in the acquired signal data. In such cases the sophisticated denoising algorithms make no sense.

Calculating accurate distance from an accelerometer during motion involves integrating its raw data and it has been well-established that when the motion is imparted by humans, consumer-grade MEMS accelerometers are rendered unsuitable for this task due to their high error-profiles even for short-interval applications. This work presents 'EMeasure', a step towards addressing this problem with a completely sensor-agnostic and elegantly accurate error-mitigating model using temporal parameters for modeling the cumulated error in acceleration and velocity, yielding accurate distance. Inherent gravity is removed using a novel latency-free method using a gyroscope. The method has been tested on stand-alone MEMS sensor boards and multiple smart devices, in both phone and wrist-watch form factor with varied IMU sensor sets. Lengths up to 5 m have been measured with a mean measurement error of less than 3 cm. As a demo, we introduce EMeasure as an immensely useful and highly accurate length-measuring utility both on smartphones and smartwatches.

This paper discusses `InLoc', an accurate and a robust positioning and tracking system using commercial mobile devices, with an integrated feature of route finding for the user from a source to the desired destination. The system exploits easily available building floor maps in raster form, with an easy conversion to vector model, eliminating the need of specially designed vector maps, thereby making the system scalable for large indoor maps and more implementation friendly. This also enables the vector map to be used in both Particle Filter based IMU tracking, and routing. Additionally, an efficient method for independent fusion of location information from phone IMU sensors and Bluetooth Low Energy (BLE) beacons is demonstrated. The method caters both dynamic and static properties of the system state. Furthermore, the paper proposes a novel approach for estimating the distance from BLE beacons using RSSI (Received Signal Strength Indication) measurement. InLoc can be readily used for any size of building floors for applications like tracking, routing and guiding system, emergency evacuation, meeting planners etc. requiring no separate effort to rebuild the vector map from scratch. The system yields a mean tracking error of less than 0.4m in location, and yields 0.9m as an average positioning error using fusion.

Recently smartphones are used every area in day-to-day life. Smartphones comes with several built-in sensors like gyroscope, accelerometer etc., along with powerful processing units. There exist various frameworks which use mobile as sensing device and mobile sensors as data extractor and process extracted data to calculate various parameter. This processing unit can be resided either in mobile side or cloud side, which provides flexibility to the researcher/developer to reduce computation time by migrating processing unit and transferring data to the cloud side. This may create problem of packet dropping or network issue while transferring data to the cloud. To overcome network issue, we propose a common framework which maintains trade-off between network overhead and processing time. The key feature of proposed framework is dividing processing unit into mobile and cloud side, sends raw data to cloud after preprocessing at mobile side. This will take very low processing time and reduce raw data size, which reduces number of packets to send to the cloud. We investigate feasibility of our proposed framework by implementing and testing with several collaborative sensing applications and comparing with the existing framework. Our result shows promising result by trading off between on-board processing and network overhead across all the solutions we had tested.

In this paper, we intend to demonstrate a novel system to measure the high-speed vibration of an anonymous vibrating object using COTS camera and optical strobing. The whole process is unobtrusive and frugal, can be used in machine inspection. The camera used has a frame rate of 30 frames per second (fps), so in conventional fashion, it is incapable to detain significant information about any vibration frequency which is not in the range of Nyquist theory of frequency (within a range of ±15 Hz). We have solved the challenge using optical strobing phenomena for capturing modulo (of division) between object's frequency and strobing frequency using camera. Motion in the video is tracked by conventional image processing technique. Finally, object vibration is calculated from the frequency plot and optical strobing frequency. Under most of the cases, the application estimates vibration frequency values, within a range of ±1.5% of error.

Photoplethysmography (PPG) signals, captured using smart phones are generally noisy in nature. Although they have been successfully used to determine heart rate from frequency domain analysis, further indirect markers like blood pressure (BP) require time domain analysis for which the signal needs to be substantially cleaned. In this paper we propose a methodology to clean such noisy PPG signals. Apart from filtering, the proposed approach reduces the baseline drift of PPG signal to near zero. Furthermore it models each cycle of PPG signal as a sum of 2 Gaussian functions which is a novel contribution of the method. We show that, the noise cleaning effect produces better accuracy and consistency in estimating BP, compared to the state of the art method that uses the 2-element Windkessel model on features derived from raw PPG signal, captured from an Android phone.

Modern day smart phones are powerful connected sensory and computation nodes for crowd-sensing, urban-sensing and personal-sensing applications. We have developed an Internet of Things (IoT) platform that can seamlessly handle data from the wide variety of sensors available on mobile phones. It can store and run aggregated analysis on the data in real-time. However, mobile phones themselves are a very heterogeneous set of devices. Each phone comes with a different array of sensors with varying sensitivity and control functions. Also, there are multiple development environments and programming languages. A final problem is seamless prototyping of applications offline and then seamless partitioning of the algorithm between phone and the cloud. In this paper we present early design elements of a framework aimed at addressing these issues.

Inertial sensing based indoor localization currently requires fairly precise layout maps, to help provide constraints and landmarks that bound the error drift. In this paper, we seek to improve the accuracy of one component of inertial-based tracking, namely the estimation of an individual's stride-length, so as to reduce the cumulative drift. We show that an individual's stride-length is affected by both his/her movement speed and heading-changes in the trajectory, and present an adaptive, online stride-length estimation algorithm that learns appropriate stride-length distributions for different (speed, heading) combinations. Initial experiments conducted using our proposed approach in combination with state-of-the-art step counting and heading estimation techniques, reduce the 95th percentile of average localization error by ≈ 30%. We thus envisage that inertial tracking may become practical even with coarse-grained map information.

Developing analytical applications for IoT based on sensor signal processing tends to be complicated as applications are executed as sequence of steps comprising of multiple alternative algorithms, including suitable feature extraction modules depending on the goal of the application. Experience shows that developers spend considerable time and effort in performing feature extraction and dimensionality reduction. In this paper we propose a framework based on a relevant case study which allows developers to drag and drop algorithms to create a workflow chain, automatically select the most relevant signal features for the particular analytic application using a training data set to generate a model and deploy the model for use. The method reduces the effort and cost of development which is deemed highly important for the analytics industry.

In this paper we propose a system for unobtrusive automated indoor surveillance of subjects in indoor environment using the Kinect sensor. We demonstrate that the features of identity, location and activity of a person can be detected with considerable accuracy using the system. Further, we show how existing design patterns can be used to create a data parallel and scalable architecture for such surveillance in real-time.

This paper describes a system for predicting water logging prone areas in multiple routes. The approach is based on the theory that water tends to accumulate in low-lying areas and hence a route which contains more and bigger basins is more likely to behave worse on a rainy day. Using this basic principle, algorithms are formulated and applied to identify and quantify water logging zones on a route. To prove the effectiveness of the proposed system, the derived confidence scores for multiple routes given by the system are compared to judgements given by human commuters. A view of all possible routes along with quantified estimates of waterlogging confidence scores are rendered in Google map.