|
| Ref# | Key factors | Merits | Demerits | Mapping | Year |
|
| [2] | EHRs (electronic health records) are supplemented with hierarchical information from medical ontologies by using a GRAM (graph-based attention model) | GRAM is performing excellently. When the data are inadequate, it works well. | Improvement is needed in the way this method incorporates knowledge DAG (directed acyclic graph) into neural networks. | RQ2 | 2017 |
|
| [7] | Restricted Boltzmann machines and autoencoder stacked units’ networks are implemented | Comparing the results of deep learning methods, which have highly precise values | Experiment with a broader scope of preprocessing methods is needed. | RQ3 | 2016 |
|
| [8] | To identify infectious disease host genes, a machine learning classification technique is developed | Wide-scale host gene prediction connected to infectious diseases is made possible. | There are no major benefits of not being able to use a small-scale dataset. | RQ1 | 2019 |
|
| [12] | Comprehensive review study for the diagnosis of COVID-19 via deep learning models | A detailed review of different diagnosis methods of COVID-19 by using the different structures of the CNN and ResNet-50 (residual network-50) model. | Computational complexity factor requires to highlight clearly | RQ2 | 2019 |
|
| [13] | Added recent research on the use of DL (deep learning) to improve the domain of health care. | Big biomedical data could be translated into improved human health by using deep learning techniques. | The development of applications needs to be improved. | RQ2, RQ3 | 2018 |
|
| [10] | CNN is a perfect model to use for the analysis of applications and challenges of medical images. | It can detect infectious disease outbreaks, among other applications. | System inconsistencies include heterogeneity of data quality and security. | RQ2 | 2021 |
|
| [14] | Use of neural networks in the prediction of diseases. | Helps to identify how neural networks can be helpful in detecting infectious diseases. | Results and technical parts are missing, which would be helpful in implementing the framework | RQ2 | 2019 |
|
| [15] | Medical, e-healthcare, and bioinformatics applications of DL are discussed. | Contains effective DL methods for biomedical and health-related applications. | In healthcare, distinctions between deep learning technologies and techniques need to be improved. | RQ2 | 2020 |
|
| [16] | SAPS II and SOFA ratings (severity scores) ML ensembles were compared for quality check. | As per the results, the DL model defeated most other techniques. | Current data must be added. | RQ2 | 2018 |
|
| [17] | Privacy concerns are highlighted in the flow of EHR through federated learning. | A unique federated learning framework proposed for efficient diagnosis of different human diseases | At least discuss the computational complexity in the flow of HER through federated learning. | RQ3 | 2018 |
|
| [18] | The fusion-based federated learning model for accurate detection of COVID-19. | Medical image analysis for detection of COVID-19 for better communication and performance if federated learning model. | Along with the accuracy factor, the robustness parameter is missed in the proposed model. | RQ3 | 2021 |
|
| [19] | Deep learning techniques are used which are working in healthcare | Exposed a few key areas of medicine where DL computational methods can have a positive impact. | Some other techniques of deep learning are not discussed | RQ2 | 2019 |
|
| [20] | Driver drowsiness is predicted by using a deep CNN model. | Helps to create an improved system that detects driver drowsiness by using the deep CNN | Needs further improvement in eye detection speed. | RQ1 | 2019 |
|
| [21] | DeepSol, a novel protein solubility predictor based on deep learning, has been proposed by researchers. | DeepSol has overcome the limitations of its feature selection step and two-stage classifier. | It can be projected with DeepSol to lower costs. | RQ2 | 2018 |
|
| [22] | FML (federated machine learning) thoroughly discusses the different parameters of training and testing the ML models. | A comprehensive review of the concepts of vertical and horizontal federated learning models. Moreover, we thoroughly discussed the applications of FML inclusion in healthcare applications. | Compromises detailed discussion on security protocols when electronic health records move from one node to another node. | RQ3 | 2019 |
|
| [23] | An evolutionary algorithm is proposed for training a DNN (deep neural network) model for the estimation of morbidity of gastrointestinal infections. | Compared to the extensively used ANN (artificial neural network) and MLR (multiple linear regression) models, this model is much more accurate at predicting disease morbidity. | Further samples should be collected, and pollutants should be determined. | RQ2 | 2017 |
|
| [24] | On the MovieQA question answering dataset, a model is presented. | Models are learning matching patterns for the selection of the right response. | To improve machine reading comprehension, the system should include entailments and answers. | RQ1 | 2018 |
|
| [25] | This study introduced the independently recurrent neural network. | By learning long-term dependencies, IndRNN (independently recurrent neural network) helps to prevent gradient explosion and disappearance. | It is not possible to improve the performance of the LSTM (long short-term memory) by raising the size of parameters or layers. | RQ1 | 2018 |
|
| [26] | The performance of ML networks is compared to that of feed-forward neural networks, also with logistic regression. | The XGB (gradient-boosted trees) model, which was found to be the most accurate, outperformed the logistic regression in terms of calibration. | There is a need for further research to improve the prediction of administrative information. | RQ1 | 2020 |
|
| [27] | The RNN technique can be formally developed for differential equations by using the RNN canonical formulation. | Signal processing-based analysis of RNNs and vanilla LSTMs and comprehensive treatment of the RNN concepts using descriptive and meaningful notation are presented. | The augmented LSTM system is effective, but it needs to be enhanced with more techniques. | RQ1 | 2020 |
|
| [28] | Developed a wearable body sensor fusion data-driven deep RNN activity recognition system. | A human’s functionality and lifestyle can be determined based on physical actions by using body sensors. | A human behaviour monitoring system can further be evaluated in real-time on overly complex datasets. | RQ1 | 2020 |
|
