Table 1.
Comparison of the related work showing the pros and cons.
| Author and year | Pros | Cons |
|---|---|---|
| 43, 2016 | Ensure the weights of the network is well optimized | |
| 62, 2016 | Breast cancer histopathological image classification | Small patches of the images are used |
| 56, 2018 | Detecting and confirming cholera and malaria epidemic pathogen | Practically limited to small datasets |
| 64, 2018 | Multi-class breast cancer classification able to predict the subclass of the tumors | Ssmaller dataset was used |
| 61, 2019 | Improved performance on image segementation | Framework was only tested on a single dataset using one set of simple network structures |
| 63, 2019 | Diagnosis of oral cancer | Limited to small amount of existing sample data |
| 24, 2020 | Self-attention mechanism on all modalities of inputs so that different anomalies features are extracted | |
| 37, 2020 | Evaluates disease severity at single time and find the site of changes in disease progression | |
| 46, 2020 | Regulate the structure of CNN to reduce the optimization time and improves the accuracy of the algorithm classification | Account the impact of different optimizers on CNN network performance not considered |
| 65, 2020 | Breast cancer histopathological image classification | Lack of pre-processing data |
| 41, 2021 | Classification of abnormality in brain MRI images | |
| 48, 2021 | Reduce image training process complexity and eliminate the over fitting problem | |
| 50, 2021 | Automatic classification of brain tumor into uncropped, cropped and segment region | Not applicable for larger image dataset |
| 51, 2021 | Enhanced image preprocessing mechanism able to detect the presence of coronavirus from digital chest X-ray | Deployed architecture incorporated parameters with high demanding memory |
| 53, 2021 | Great choice for accurate delineation of tumor margin | Both object detection and segmentation belongs to supervised algorithms which required experienced doctors to label images |
| 54, 2021 | Detection model for coronavirus using CT and X-ray image data | The model was a theoretical framework which was not subjected nor verified in actual clinical practices |
| 57, 2021 | Cclassified the medical images based on anatomic location and modality | Images were limited to JPEG, no preprocessing medium for images and it was subjected to small dataset |
| 68, 2021 | Clinical information to predict pathology complete response (pCR) to neoadjuvant chemotherapy (NAC) | |
| 39, 2022 | Leverages on the benefit of few-shot learning, to address the problem of detecting COVID-19 CT scan images | |
| 59, 2022 | Multicolor imaging for the purpose of extracting features which reveals sufficient symptoms to arrive at the detection of diseases | |
| 40, 2022 | Solved classification problem on hyperspectral images to exploit spatial context and spectral bands jointly | |
| 44, 2022 | Minimize the challenge of insufficient training dataset | |
| 46,2022 | High classification performance in breast cancer detection from mammography images | May not be generalized to other pretrained CNN architecture and limited to a specific dataset |
| 47, 2022 | Iris image recognition | The performance measures of the proposed methods are limited to the IIT Delhi database and the performance of the network may fail for other iris databases |
| 49, 2022 | Demonstrated and classified architectural distortion, assymetric and macro-calcification abnormalities | Proposed architecture not compatible to whole image |
| 58, 2022 | Wounds and their location in the body using multimodal approach | |
| 28, 2023 | Feature extraction and fusion on multimodal images to support the classification accuracy and localization of medical images | |
| 29, 2023 | Eliminate noise and distortion in data stream associated with electrocardiography | |
| 33, 2023 | Eliminate erroneous predictions | |
| 52, 2023 | Demonstrates the advantage of combining mammography images and clinical data | Only small dataset was used to train the model |