Table 7 Comparison of proposed 6G/WiFi integrated sensing and communication (ISAC) passive sensing-based contactless pulmonary disease screening method with state-of-the-art techniques

From: Non-contact lung disease classification via orthogonal frequency division multiplexing-based passive 6G integrated sensing and communication

Reference

Sensing/ dataset used

Model

Lung/ respiratory diseases identified

Participants

Performance

Remarks

Malik et. al.13

X-ray images

VGG-19, ResNet-50, Inception v3, CDC_Net

COVID-19, Lung Cancer, Pneumothorax, TB, Pneumonia

COVID-19: 2371, Pneumothorax: 12,000, Pneumonia: 3867, Lung Cancer: 5000, TB: 4200, Normal: 2949, Total: 30387

VGG-19: 95.61%, ResNet-50: 96.15%, Inception v3: 95.16%, CDC_Net: 99.39%

Proposed CDC-Net, a CNN with residual connections and dilated convolutions to enhance classification accuracy.

Tuncer et. al.14

X-ray images

DT, LD, SVM, Ensemble, k-NN

COVID-19, Pneumonia, Normal

COVID-19: 135, Pneumonia: 150, Normal: 150, Total: 435

DT: 86.09%, LD: 82.30%, SVM: 97.01%, k-NN: 96.09%, Ensemble: 93.10%

616 features extracted from X-rays to improve the classification and classified with six ML models.

Mei et. al.18

CT scans

CNN

ILD and Normal

ILD: 458

Sensitivity: 82.4%

Classified ILD and predicted progression and prognosis using CT scan data.

Choi et. al.22

Respiratory Sound Database

Light Attention, VGGish

Bronchiectasis, COPD, ILD, Pneumonia, Asthma, Normal

Total: 126 (Normal: 12, Asthma: 23, COPD: 20, ILD: 26, Pneumonia: 20, Bronchiectasis: 25)

Overall: 92.81%, Class-wise accuracy between 89 to 97%

Used electronic stethoscope for data collection; introduced LACNN with XAI for interpretability.

Xu et. al.27

Cough based Database

MobileNet

COPD, Asthma, Chronic cough, Normal

Total: 201 (Normal: 50, Asthma: 94, COPD: 34, Asthma + COPD: 13, Chronic cough: 10)

94.8

built an end-to-end audio-based cough detection model using a pre-trained MobileNet and a variety of audio augmentation techniques.

Siddiqui et. al.34

UWB radar

LSTM, SVM, KNN, Decision Tree

COPD and Healthy

70 (35 COPD, 35 healthy)

LSTM: 96.3%, SVM: 94.5%, KNN: 92.7%, DT: 90.9%

Radar-based contactless system for COPD detection.

Beltrao et. al.40

CW radar dataset

Nonlinear Least Squares (NLS) Estimator

Normal and Cheyne-Stokes

12 premature infants

97%

Monitored respiratory patterns of premature infants using radar

Rehman et. al.45

SDR based CSI

Cosine KNN, Complex Tree, Boosted Tree, SVM

Eupnea, Bradypnea, Tachypnea, Biot, Sighing, Kussmaul, Cheyne-Stokes, CSA

5 subjects, 8 patterns, 10 repetitions = 400 samples

Cosine KNN: 97.5%, Complex Tree: 96.8%, Boosted Tree: 85.6%, SVM: 75.5%

Non-invasive recognition of 8 abnormal patterns.

Kawish et. al.46

SDR based CSI

K-NN, SVM

Eupnea, Tachypnea, Bradypnea

4 subjects, 3 breathing pattern, 5 repetitions = 60 experiments total

Chest based performance: 95.9%, Hand based performance: 88.1%

Monitored breathing patterns using SDR-based CSI reflected from the chest, and also attempted to determine these patterns from reflections off the hand.

Proposed method

OFDM-based SDR data

MLP (Medium)

Five lung diseases + Healthy

190 + 30

95.9%

Explained in section III-B

Proposed method

OFDM-based SDR data

CNN

Five lung diseases + Healthy

190 + 30

98%

Explained in section III-B

Back to article page