Disease Prediction Using Machine Learning on Smartphone-Based Eye, Skin, and Voice Data: Scoping Review

JMIR AI, Journal Year: 2025, Volume and Issue: 4, P. e59094 - e59094

Published: March 25, 2025

Background The application of machine learning methods to data generated by ubiquitous devices like smartphones presents an opportunity enhance the quality health care and diagnostics. Smartphones are ideal for gathering easily, providing quick feedback on diagnoses, proposing interventions improvement. Objective We reviewed existing literature gather studies that have used models with smartphone-derived prediction diagnosis anomalies. divided into those conducting experiments retrieve predict diseases, publicly available databases. details databases, experiments, intended help researchers working in fields artificial intelligence domain. Researchers can use information design their or determine databases they could analyze. Methods A comprehensive search PubMed IEEE Xplore was conducted, in-house keyword screening method filter articles based content titles abstracts. Subsequently, related 3 areas voice, skin, eye were selected analyzed how extracted (ie, through experiments). each study also noted. Results total 49 identified as being relevant topic interest, among these studies, there 31 different 24 methods. Conclusions results provide a better understanding smartphone collected predicting diseases what kinds data. Similarly, having smartphone-based be various been presented. Our improved future our findings reference conduct similar statistical analyses.

Language: Английский

---

A Performance Study of CNN Architectures for the Autonomous Detection of COVID-19 Symptoms Using Cough and Breathing

Computers, Journal Year: 2023, Volume and Issue: 12(2), P. 44 - 44

Published: Feb. 17, 2023

Deep learning (DL) methods have the potential to be used for detecting COVID-19 symptoms. However, rationale which DL method use and symptoms detect has not yet been explored. In this paper, we present first performance study compares various convolutional neural network (CNN) architectures autonomous preliminary detection of cough and/or breathing We compare analyze residual networks (ResNets), visual geometry Groups (VGGs), Alex (AlexNet), densely connected (DenseNet), squeeze (SqueezeNet), identification ResNet (CIdeR) investigate their classification performance. uniquely train validate both unimodal multimodal CNN using EPFL Cambridge datasets. Performance comparison across all modes datasets showed that VGG19 DenseNet-201 achieved highest DensNet-201 had high F1 scores (0.94 0.92) on dataset, compared next score (0.79), with comparable larger dataset. They also consistently accuracy, recall, precision. For detection, (0.91) other structures (≤0.90), having accuracy recall. Our investigation provides foundation needed select appropriate deep utilize non-contact early detection.

Language: Английский

---

BreathPro: Monitoring Breathing Mode during Running with Earables

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies, Journal Year: 2024, Volume and Issue: 8(2), P. 1 - 25

Published: May 13, 2024

Running is a popular and accessible form of aerobic exercise, significantly benefiting our health wellness. By monitoring range running parameters with wearable devices, runners can gain deep understanding their behavior, facilitating performance improvement in future runs. Among these parameters, breathing, which fuels bodies oxygen expels carbon dioxide, crucial to improving the efficiency running. While previous studies have made substantial progress measuring breathing rate, exploration additional during still lacking. In this work, we fill gap by presenting BreathPro, first mode system for It leverages in-ear microphone on earables record sounds combines out-ear same device mitigate external noises, thereby enhancing clarity sounds. BreathPro incorporates suite well-designed signal processing machine learning techniques enable detection superior accuracy. We implemented as smartphone application demonstrated its energy-efficient real-time execution.

Language: Английский

---

Novel Speech-Based Emotion Climate Recognition in Peers’ Conversations Incorporating Affect Dynamics and Temporal Convolutional Neural Networks

Published: Jan. 1, 2024

Peers' conversation provides a domain of rich emotional information, conveyed not just through facial expressions and gestures, but also their speech itself. This ongoing exchange creates dynamic climate (EC) that influences social interaction behavior, offering valuable insights beyond the content words.Recognition EC could provide an additional source in understating peers' behavior on top actual conversational content.Here, we propose novel approach for speech-based recognition, namely AffECt, by combining complex affect dynamics (AD) with deep features extracted from signals using Temporary Convolutional Neural Networks (TCNNs). AffECt was tested cross-validated data drawn three open datasets, i.e., K-EmoCon, IEMOCAP, SEWA, terms arousal/valence level classification. The experimental results have shown achieves classification accuracy up to 83.3\% 80.2\% arousal valence, respectively, clearly surpassing reported literature, exhibiting robust performance across different languages. Moreover, there is distinct improvement when AD are combined TCNN, compared baseline learning approaches. These demonstrate effectiveness paving way many applications, e.g., patients' group therapy, negotiations, emotion-aware mobile applications.

Language: Английский

---

Smartphone-based device for point-of-care diagnostics of pulmonary inflammation using convolutional neural networks (CNNs)

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 22, 2024

Abstract In pulmonary inflammation diseases, like COVID-19, lung involvement and determine the treatment regime. Respiratory is typically arisen due to cytokine storm leakage of vessels for immune cells recruitment. Currently, such a situation detected by clinical judgment specialist or precisely chest CT scan. However, lack accessibility machines in many poor medical centers as well its expensive service, demands more accessible methods fast cheap detection inflammation. Here, we have introduced novel method tracing patients with inflammation, simple electrolyte their sputum samples. The presence sample results fern-like structures after air-drying. These fern patterns are different positive negative cases that an AI application on smartphone using low-cost portable mini-microscope. Evaluating 160 patient-derived images, this demonstrated interesting accuracy 95%, confirmed CT-scan results. This finding suggests has potential serve promising reliable approach recognizing inflammatory COVID-19.

Language: Английский

---

Deep learning identifies cardiac coupling between mother and fetus during gestation

Frontiers in Cardiovascular Medicine, Journal Year: 2022, Volume and Issue: 9

Published: July 29, 2022

In the last two decades, stillbirth has caused around 2 million fetal deaths worldwide. Although current ultrasound tools are reliably used for assessment of growth during pregnancy, it still raises safety issues on fetus, requires skilled providers, and economic concerns in less developed countries. Here, we propose deep coherence, a novel artificial intelligence (AI) approach that relies 1 min non-invasive electrocardiography (ECG) to explain association between maternal heartbeats pregnancy. We validated performance this using trained learning tool total 941 one minute maternal-fetal R-peaks segments collected from 172 pregnant women (20-40 weeks). The high accuracy achieved by (90%) identifying coupling scenarios demonstrated potential AI as monitoring frequent evaluation development. interpretability was significant explaining synchronization mechanisms heartbeats. This study could potentially pave way toward integration automated clinical practice provide timely continuous while reducing triage, side-effects, costs associated with devices.

Language: Английский

---

Analyzing the effect of data preprocessing techniques using machine learning algorithms on the diagnosis of COVID‐19

Concurrency and Computation Practice and Experience, Journal Year: 2022, Volume and Issue: 34(28)

Published: Oct. 18, 2022

Real-time polymerase chain reaction (RT-PCR) known as the swab test is a diagnostic that can diagnose COVID-19 disease through respiratory samples in laboratory. Due to rapid spread of coronavirus around world, RT-PCR has become insufficient get fast results. For this reason, need for methods fill gap arisen and machine learning studies have started area. On other hand, studying medical data challenging area because it contains inconsistent, incomplete, difficult scale, very large. Additionally, some poor clinical decisions, irrelevant parameters, limited adversely affect accuracy performed. Therefore, considering availability datasets containing blood which are less number than today, aimed improve these existing datasets. In direction, obtain more consistent results studies, effect preprocessing techniques on classification was investigated study. study primarily, encoding categorical feature scaling processes were applied dataset with 15 features contain 279 patients, including gender age information. Then, missingness eliminated by using both K-nearest neighbor algorithm (KNN) equations multiple value assignment (MICE) methods. Data balancing been done synthetic minority oversampling technique (SMOTE), method. The ensemble algorithms bagging, AdaBoost, random forest popular classifier KNN classifier, support vector machine, logistic regression, artificial neural network, decision tree classifiers analyzed. highest accuracies obtained bagging 83.42% 83.74% MICE imputations applying SMOTE, respectively. ratio reached same without SMOTE 83.91% imputation. conclusion, certain examined comparatively success presented importance right combination achieve demonstrated experimental studies.

Language: Английский

---

Cloud-based COVID-19 disease prediction system from X-Ray images using convolutional neural network on smartphone

Multimedia Tools and Applications, Journal Year: 2022, Volume and Issue: 82(19), P. 29883 - 29912

Published: Nov. 24, 2022

Language: Английский

---

Efficient Pandemic Infection Detection Using Wearable Sensors and Machine Learning

2022 International Wireless Communications and Mobile Computing (IWCMC), Journal Year: 2023, Volume and Issue: unknown, P. 1562 - 1567

Published: June 19, 2023

More than three years into the coronavirus disease 2019 (COVID-19) pandemic, it can be noted that measures put in place for societies to manage spread of this could have been better. For example, contact tracing mobile applications used curb COVID-19 need additional enhancements allow health care professionals better understand proliferation and lessen burden on hospitals medical centers. In paper, we present an intelligent solution remotely self-monitor symptoms help rapidly identify detect suspected positives. The proposed is based using a near-field communications (NFC) wristband collects body temperature heart rate SpO2 levels. It connected dedicated application intelligently draw conclusions from data (COVID-19 symptoms) collects. Moreover, trained analyze cough sounds probability infection. Results show more 90% detection accuracy. system adapted future pandemics respiratory symptoms.

Language: Английский

---

A Development of a Sound Recognition-Based Cardiopulmonary Resuscitation Training System

IEEE Journal of Translational Engineering in Health and Medicine, Journal Year: 2024, Volume and Issue: 12, P. 550 - 557

Published: Jan. 1, 2024

The objective of this study was to develop a sound recognition-based cardiopulmonary resuscitation (CPR) training system that is accessible, cost-effective, easy-to-maintain and provides accurate CPR feedback. Beep-CPR, novel device with accordion squeakers emit high-pitched sounds during compression, developed. emitted by Beep-CPR were recorded using smartphone, segmented into 2-second audio fragments, then transformed spectrograms. A total 6,065 spectrograms generated from approximately 40 minutes data, which randomly split training, validation, test datasets. Each spectrogram matched the depth, rate, release velocity compression measured at same time interval ZOLL X Series monitor/defibrillator. Deep learning models utilizing as input trained transfer based on EfficientNet predict depth (Depth model), rate (Rate (Recoil model) compressions. Results: mean absolute error (MAE) for Depth model 0.30 cm (95% confidence [CI]: 0.27-0.33). MAE Rate 3.6/min CI: 3.2-3.9). For Recoil model, 2.3 cm/s 2.1-2.5). External validation demonstrated acceptable performance across multiple conditions, including utilization newly-manufactured device, fatigued evaluation in an environment altered spatial dimensions. We have developed system, accurately measures quality training. Significance: cost-effective solution can improve efficacy facilitating decentralized at-home

Language: Английский

---