Table 6.

IoT/IoMT-based abnormality and arrhythmia detection.

Wearable/Smart/Medical Device	Approach	Results
VA Processor/SoC, (Custom-made) [47]	Naive Bayes	Accuracy: 86%, Power consumption reduction: 62.2%
Electrodes (3 unipolar limb leads, 3 bipolar limb leads, 6 unipolar chest leads) [48]	Convolutional Neural Network	Accuracy: 98%, Sensitivity: 96%
Lenovo Smart ECG vest, (Lenovo Group Ltd., Beijing, China) [49]	Convolutional Neural Network	Accuracy: 86.3%
Arduino Uno, (Arduino, Scarmagno, Italy), Raspberry Pi 3B, (Raspberry Pi Foundation, Cambridge, UK) AD8232 ECG sensor (DFRobot, Shanghai, China), [50]	k-NN	Accuracy: 94.44%
Biomedical sensors, ARM processor, FPGA [51]	k-NN	Accuracy: 99%
Intelligent electrocardiograph device [52]	Neural network architecture based on deep learning	1st network Accuracy: 91%, 2nd network Accuracy: 100%, 3rd network Accuracy: 90%
AD8232 EKG sensor, (SparkFun Electronics, Niwot, CO, USA), Arduino board, (Arduino, Scarmagno, Italy), Jetson Nano microcomputer, (Nvidia Corporate, Santa Clara, CA, USA) [53]	Dynamic mode selected energy, adaptive window sizing, R location correction algorithm for detecting R-peaks with better efficiency	Accuracy: 99.94%, Sensitivity: 99.98%, Precision: 99.96% Specificity: 99.98% AUC: 99.89% Detection error rate: 0.06%
Raspberry Pi 3B (Raspberry Pi Foundation, Cambridge, UK) [54]	Fourier Transform, Convolutional Neural network (CNN)	Accuracy: 99.91% F1-Score: 95% Average inference time: 9 ms Maximun memory usage: 12 mb%
SensorTile (STEVAL-STLKT01V1), (STMicroelectronics, Grenoble, France), AD8232 (DFRobot, Shanghai, China), Raspberry Pi (Raspberry Pi Foundation, Cambridge, England, UK) [55]	Convolutional Neural network (CNN)	Accuracy: 97% Sensitivity: 96.92% Precision: 91.50% F1-Score: 94.89%
Raspberry Pi 4 (Raspberry Pi Foundation, Cambridge, UK) [56]	1D Convolutional Neural network (1D-CNN) GridSearch	Accuracy: 99.46%
Arduino Uno, (Arduino, Scarmagno, Italy), ATMEGA328P Microcontroller, (Microchip, AZ, USA) Raspberry Pi (Raspberry Pi Foundation, Cambridge, UK) [57]	Incremental Support vector Regression	Accuracy: 98.5% Sensitivity: 88% Precision: 90% Specificity: 99%
Sensor nodes [58]	Convolutional Neural network (CNN)	Accuracy: 95% Sensitivity: 94.63% Specificity: 94.63% ROC: 96.53%
Diagnosis and Tracking Shield, (Custom-made), ADS1298 (TX Instruments, Dallas, TX, USA), Raspberry Pi (Raspberry Pi Foundation, Cambridge, UK) [59]	Depth Convolutional Neural Network	Accuracy: 96.67% Sensitivity: 96.63% Specificity: 96.67%
ECG Machine [60]	Convolutional Neural network (CNN)	Accuracy: 99.12% Sensitivity: 100% Specificity: 99.12%
Smartphone device [61]	Convolutional Neural network (CNN)	Accuracy: 93%
MAC 5500 HD, (GE Healthcare, Chicago, IL, USA), MUSE v9, (GE Healthcare, Chicago, Illinois, USA) [62]	Convolutional Neural network (CNN)	Sensitivity: 88.50% Specificity: 88.54% Positive Predictive: 88.54% Negative Predictive: 88.54% F1-Score: 88.49%
Wearable sensors [63]	Convolutional Neural network (CNN), Artificial Bee Colony, Grey Wolf Optimizer	Accuracy: 94% Recall: 94.5% Precision: 96% Specificity: 95.4%
Noninvasive healthcare sensor, SkopEdge (Custom-made, India), Raspberry Pi, (Raspberry Pi Foundation, Cambridge, UK) [64]	Randon Forest	MIT-BIH Accuracy: 98.53% PTB Accuracy: 99% RF Accuracy: 98.68%
BH1790GLC (Rohm, Kyoto, Japan) [65]	Convolutional Neural network (CNN)	Sensitivity: 99.5% Specificity: 98.7% F1-Score: 99.1% Time: 19 s%
Sony Xperia Z-series, (Sony, Tokyo, Japan) [66]	Kernel SVM	Accuracy: 97.4%, Sensitivity: 93.8%, Specificity: 100%
AFE4403 (TX Instruments, Dallas, TX, USA) [67]	Linear Kernel SVM	TPR ${}^1$: 70.10%, TNR ${}^2$: 88.61%, Accuracy: 80.37%
Mason-Likar ECG 12-lead system (CardioCloud Medical Technology, Beijing, China) [68]	Deep Densely Connected Neural Network (DDNN)	Accuracy: 96.73%, Sensitivity: 96.67%, Specificity: 96.93%
KardiaMobile EKG Monitor (AliveCor Inc., CA, USA) [69]	Neural Network	AUC ${}^3$: 82.7%, Specificity: 74.9%
Sony Xperia Z1/Z5, (Sony, Tokyo, Japan), Philips IntelliBue MX40 (Philips, Amsterdam, Netherlands) [70]	Random Forest, XGBoost, Logistic Regression	AUC AFib ${}^4$: 98%, 98%, 96%, AUC ADHF${}^5$: 85%, 82%, 83%
ZYNQ Ultrascale ZCU106 FPGA, (Advanced Micro Devices, Inc., Santa Clara, CA, USA) [71]	1D Convolutional Neural network (1D-CNN)	Accuracy: 99.17%, Sensitivity: 97.03%, Specificity: 99.37%, Precision: 93.72%, F1-score: 97.90%

¹ True positive rate. ² True negative rate. ³ Data obtained from [124]. ⁴ Atrial fibrillation. ⁵ Acute decompensated heart failure.