|
| Source | Purpose | Major findings | Accuracy (%) |
|
| Azar et al. [10] | Detection of vertebral column pathology | A decision support tool is proposed for the detection of pathology on the vertebral column using three types of decision trees classifiers. | SDT: 81.94
BDT: 84.84
DTF: 84.19 |
|
| Indriana et al. [11] | Classification of vertebral column disease | Ensembled decision tree (J48) and bagging are used as the classification model. | Ensemble model: 85%
J48: 81% |
|
| Choudhary et al. [16] | IDC classification | Pruned models performed superior over original pretrained models. | 92.07 |
|
| Samala et al. [17] | Breast cancer diagnosis | To compress a deep convolutional neural network for mass classification in digital breast tomosynthesis (DBT), a layered pathway evolution strategy is presented. | — |
|
| Hu et al. [18] | Network trimming | The zero activation neurons are unnecessary and may be eliminated without impacting the network’s overall accuracy. | 90.278 |
|
| Hajabdollahi et al. [19] | Retinal disease screening and diagnosis | A simplification approach is proposed for CNNs based on the combination of quantization and pruning. | 76 |
|
| Hajabdollahi et al. [20] | Detection and analysis of diabetic retinopathy | To simplify the network topology, hierarchical pruning gradually removes connections, filter channels, and filters. | 92 |
|
| Mantzaris et al. [21] | Medical disease prediction | This research uses a genetic algorithm (GA) to prune probabilistic neural networks. | 85.5 |
|
| Yin et al. [22] | Diabetes diagnosis | DiabDeep is a paradigm for widespread diabetes detection that blends efficient neural networks (named DiabNNs) with off-the-shelf WMSs. | 94 |
|
| Chen and Zhao [23] | Reducing complex CNNs | The pruning procedure is conducted at the layer level, and redundant parameters were discovered by studying the features learned in the convolutional layers. | 93.03 |
|
| Han et al. [24] | Deep compression | A three-stage pipeline approach (pruning, trained quantization, and Huffman coding) is introduced to reduce the storage requirement of neural networks. | — |
|
| Li et al. [25] | Pruning and compressing | Filters from CNNs that have been recognized as having a little impact on output accuracy have been pruned. | — |
|
| Horry et al. [26] | Lung cancer diagnosis | An improved generalization can be achieved with an image preprocessing pipeline that homogenizes and debases chest X-ray images and helps to develop a low-cost, accessible DL system for lung cancer screening. | 89 |
|
| Xiang et al. [27] | Skin disease diagnosis | Without changing the model size, the performance of the model is improved after fine-tuning. | 83.5 |
|
