Table 2.
Review of DL applications in medical image processing.
| Ref. | Dataset | Topic | Methods used | Research challenges |
|---|---|---|---|---|
| Lo et al. (1995) | 55 chest X-ray images | Lung Nodule Classification | Artificial CNN | Texture assessment approaches to identify disease trends |
| Shen et al. (2015) | Lung Image Database Consortium | Lung Nodule Classification | Multi-scale CNN | The initial training and testing set strikingly different |
| Rajpurkar et al. (2017) | 1,12,120 X-ray images of 30,805 patients. | Pathological classification of pneumonia | CheXNet DL model | Considered F1-Score only as performance metrics |
| Pratt et al. (2016) | Kaggle dataset | Classification for DR | CNN | System failed to learn more complex features |
| Abràmoff et al. (2016) | Messidor-2 Â- ADCIS dataset | Classification for DR | CNN+ IDx-DR X2.1 | Failed to substitute CNN-trained features |
| Kawahara and Hamarneh (2016) | Private Skin dataset | Classification of skin lesions | Multi-layer CNN | Excluded use of bigger skin dataset |
| Roth et al. (2015) | 4298 X-ray of 1675 patients | Classification of organs | CNN | Predictions are sluggish, allowing implementation problems |
| Guo et al. (2015) | University of Chicago Hospital. | Localization of prostate | SSAE | Considered only 66 images of prostate |
| Shin et al. (2012) | DCE-MRI dataset | localize multi-organ disease | Single-Layer SSAE | Limited dataset, System failed to learn more complex features |
| Payer et al. (2016) | Private Dataset | Accurate response with landmark localization of the medical image | SCN Architecture | Strategies to minimize the complexity of the system are not included. |
| Baumgartner et al. (2016) | 1003 pregnancy scan reports | Localize the fetal | CNN | All the performance metrics not evaluated with traditional models |
| Ghesu et al. (2016b) | 869 patients, 2891 aortic valve images | Object detection | Marginal Space DL | Failed to address computational constraints |
| Shin et al. (2016) | 905 images, 120 patients | Detect interstitial lung disease | Deep CNN | Failed to deal with theoretical work on cross-modality statistics |
| Liao et al. (2019) | Kaggle dataset | Lung cancer detection | 3-D NN | Failed to detect high accuracy for small nodules |
| Xu et al. (2015) | Case Western Reserve University | Detection of nuclei in breast images | SSAE | Requires improvement in extraction of features |
| Cruz-Roa et al. (2013) | 1417 skin images | Detect cancer in the skin | Softmax classifier | Excluded use of bigger skin dataset |
| Guo et al. (2019) | Soft Tissue Sarcoma dataset | Tumor segmentation | Deep CNN | Examined a single dataset on a single network. |
| Oktay et al. (2017) | UK Digital Heart Project dataset | Cardiac image segmentation | CNN | Low resolution slice |
| Seeböck et al. (2019) | 226 images, 33 healthy volumes | Retinal anatomy segmentation | CNN | Requires improvement in learning process |
| Zhu et al. (2019) | 81 prostate MR volumes | Prostate image segmentation | BOWDA-Net | Limited dataset, system failed to learn more complex features |
| Elmahdy et al. (2019) | Haukeland Medical Center cancer dataset | Cancer registration | Elastix automated 3D deformable registration software | Strategies to minimize the complexity of the system are not included. |
| Bai et al. (2013) | Short-axis cardiac magnetic resonance data set | Cardiac registration | Multi-atlas classifier | Requires better computational capability |
| Chee and Wu (2018) | Private MR brain image dataset | 3D-image registration | Self-supervised learning model | Findings are confined to brain scans with axial vision |
| Lv et al. (2018) | 27 healthy members | Detect motion-free abdominal images | CNN image registration model | Tiny dateset with minor lesions |