Table 8 Previous studies for PE classification. Here, AUC, CC, MIL, sen, spc, SSL, SPE-YOLO stand for area under the curve, conventional classification, multiple instance learning, sensitivity, specificity, self-supervised learning, and SE-Attention Prioritizes Features PE-You Only Look-Once respectively.
Study | Dataset | Architecture | Task Type | Findings |
|---|---|---|---|---|
Condrea et al.15 | RSNA | Dual-hup DNN with anatomical aware pretraining | PE detection from CT images | sen = 92.05%, spc = 96.17% |
Ma et al.26 | RSNA | Two phase multi task learning with interpretability(Grad-CAM, attention) | PE detection +localization +chronicity+ RV/LV ratio | AUROC = 0.93, sen = 86.02% |
Khan et al.27 | RSNA | DL framework based on DenseNet201 (feature extractor)+ customized fully connected layers | Multi-classification (PE classification into 9 classes) | Acc = 88.01%, sen = 88.00%, AUC = 0.90 |
Islam et al.28 | CTPA dataset (specific dataset not mentioned) | Comparative study: CNNs vs ViTs; SSL vs supervised; transfer learning vs training from scratch; CC vs MIL | PE diagnosis(image-level and exam-level) | AUC = 0.96 |
Lynch et al.32 | RSNA | PE-DeepNet: hybrid deep CNN with reduced parameters | PE classification | Acc = 94.21% |
Suman et al.36 | RSNA | Two stage attention-based CNN-LSTM network | PE detection+ type(chronic/acute)+ location(left/right/central) | AUC = 0.95 |
Wu et al.35 | Tianjin internal(n=142) +RSNA test set(n=2000) | SPE-YOLO: YOLOv8 + P2 head+ SE-Attention+ ODconv for small PE detection | Small PE detection | sen = 90.71%, Acc = 86.45% |
Mohammed et al.20 | RSNA | EfficientNet-B7 + enhanced ViT with multi-task learning | Multi-task classification(PE detection, location, type) | AUC = 0.96 |
Cahan et al.37 | Internal multimodal dataset(3D CTPA + clinical data) | Bilinear attention+ TabNet(structured + maging) | PE severity risk stratification (classification) | AUC = 0.96, sen = 90.00%, spc = 94.00% |
The proposed method | RSNA | Ensemble approach (ResNet50 + DenseNet121 + Swin Transformer) | Binary classification | Acc = 97.80%, AUROC = 0.99 |
