Table 1 Promising developments and persistent obstacles for kidney disease classification.

Transfer learning techniques

They used pre-trained models (e.g., ResNet, VGG) to enhance the accuracy of kidney disease classification

It does not generalize well to all datasets; performance heavily depends on the source dataset quality

^2,5,41

Vision transformer models

Use modified Vision Transformer (ViT) models to classify CT scan images effectively

It requires large datasets for practical training and may be less interpretable than traditional CNNs

^16,27

Fine-Tuning Techniques

Fine-tuning pre-trained models to adapt them to specific tasks improves diagnostic performance

It requires careful selection of hyperparameters; it can lead to overfitting if not appropriately managed

^16,42

Hyperparameter optimization

Advanced methods like Bayesian optimization and genetic algorithms to optimize model parameters

Computationally intensive; may require extensive resources and time for large datasets

^14,29

Fruit fly optimization algorithm

They utilized optimization algorithms for feature selection in CKD classification

It may be sensitive to parameter settings and requires careful tuning to achieve optimal results

^14,29

Hybrid models

The authors integrated different architectures (e.g., CNNs with LSTMs) for improved prediction capabilities

Complexity in training and tuning may require large amounts of labeled data for practical training

^15,28

Ensemble learning approaches

They combined multiple models to enhance classification accuracy and robustness

Increased complexity in model management; potential for longer inference times

^25,40,43

Deep semantic segmentation models

The authors utilized deep semantic segmentation for kidney segmentation and stone detection

Image noise can affect performance and may require extensive labeled data for training

^11,12

Cycle-GAN for segmentation

They utilized Cycle-GAN for semi-supervised kidney segmentation, achieving high accuracy

Limited to specific imaging modalities; requires substantial computational resources for training

^6,7

Segmentation and detection approaches

Enhanced kidney segmentation using deep learning techniques, improving the detection of lesions and stones

Limitations in detecting small lesions: segmentation accuracy can vary based on image quality

^7,9

Multi-classification approaches

Use of multi-classification models to categorize various renal diseases effectively. The dual-augmentation method is used to achieve high accuracy

They may struggle with imbalanced datasets and require careful handling of class distribution in training

^14,42,20,21

Automated CNN classification approaches

The researchers developed a CNN-based system for automated detection of kidney stones in ultrasound images

Performance may degrade with lower-quality images and limited generalization across different imaging modalities

^17,40

Fuzzy deep neural networks

The researchers incorporated fuzzy logic in deep learning models for improved prediction of kidney diseases

Complexity in model design may require extensive tuning to achieve optimal performance

^25,26

Morphological cascade CNNs

Innovative use of morphological cascade CNNs for enhanced detection of small kidney lesions

Complexity in architecture design may require extensive training data for effective learning

^29,28