Table 8 Key characteristics and performance of SOTA for kidney disease classification.
From: An attention enhanced dilated bottleneck network for kidney disease classification
Ref | Algorithm | Method | Key characteristic | Accuracy (%) |
|---|---|---|---|---|
UNet, SegNet, MobileNetV2, VGG16, and InceptionV3 | Transfer Learning | Manual annotation of kidney CT images with expert validation and Grad-CAM visualisation for explainable model decisions | (VGG16) 99.48 | |
DenseAUXNet201 | Transfer Learning | Principal Component Analysis (PCA) to reduce dimensionality and feature selection techniques for optimisation. | 90.63 | |
XResNet 50 with SHAP | Transfer Learning | An early stopping function is employed to prevent overfitting. | 97 | |
AHDCNN and SVM | Deep Convolutional Neural Network | The model used a multi-layer deep belief network to identify irregular patterns in kidney disease data. | 97.3 | |
Deep Learning Neural Network, Inception V3 | DLNN, Transfer Learning Model | The architecture balances depth and computational complexity. | 91 | |
VGG19 and Naive Inception | Transfer Learning Model | Additional layers were used to address the issues of vanishing gradients and overfitting. | 99.25 | |
RBCA-Net, ResUNet | Transfer Learning | Atrous Spatial Pyramid Pooling (ASPP) is utilised to extract spatial information. | 82.79 | |
Coarse to fine kidney segmentation | Convolutional Neural Network | Abnormality detection is performed using component analysis and a 2D convolutional neural network to correct the abnormal regions. | 98.69 | |
Proposed model | Dilated Bottleneck Attention-based Renal Network (DBAR-Net) Proposed Model | Multi-Feature Fusion Technique | Two fold convolved layer normalization blocks, dual bottleneck attention modules, dilated convolved layer normalisation block | 98.86 |
