Table 4 Comparison of advanced models for the 3D reconstruction of lung tumors.
Authors | Methodology | Contribution | Dataset | Result | Limitation |
|---|---|---|---|---|---|
Hong et al.10 | VGG and GK clustering | Introduces GAN-based 3D reconstruction for lung tumor imaging | LUNA16 | HD: 2.82 and ED: 2.82 | Performance may vary for non-lung tumor datasets |
Gu et al.83 | ResNet and GK clustering | Combines GAN and LSTM for feature-driven 3D tumor reconstruction | LUNA16 | HD: 2.99 and ED: 1.06 | Requires well-segmented 2D input images |
Rezaei et al.84 | VGG and GK clustering | Uses GAN for 3D reconstruction and transfers 2D image features | LUNA16 | HD: 3.02 and ED: 1.06 | Relies heavily on accurate segmentation in preprocessing |
Karrar et al.86 | 3D reconstruction using bounding boxes and rule-based classifier | Develops an approach for nodule extraction and surface rendering for 3D modeling | LIDC-IDRI | Accuracy: 99.6627% | Limited robustness to nodules with irregular shapes |
Dlamini et al.87 | YOLOv4 and region-based active contour model | Integrates detection and volumetric rendering in a single automated pipeline | LIDC-IDRI | Accuracy: 99.74% and DSC: 92.19% | Requires robust preprocessing for noise removal |
Shi et al.88 | ViT and demographics | Enhances 3D reconstruction from 2D images using ViT | 2525 chest x-rays | DSC: 76.9% | Dependent on the inclusion of accurate demographic data |
Najafi et al.2 | GAN, LSTM, and VGG16 | Integrates GANs with VGG16 and LSTM for accurate 3D reconstruction | LIDC-IDRI | HD: 0.986 and ED: 1.126 | Training instability due to sensitivity to input variations and multi-model tuning |
Huang et al.89 | GAN, attention-based LSTM, and VGG16 | Integrates multi-stage GANs with VGG16 and attention-based LSTM for accurate 3D reconstruction | LIDC-IDRI | HD: 2.963 and ED: 1.725 | Requires extensive labelled data and balanced segmentation to maintain accuracy across GAN stages |
