An AI healthcare ecosystem framework for Covid-19 detection and forecasting using CronaSona

The primary purpose of this paper is to establish a healthcare ecosystem framework for COVID-19, CronaSona. Unlike some studies that focus solely on detection or forecasting, CronaSona aims to provide a holistic solution, for managing data and/or knowledge, incorporating detection, forecasting, expert advice, treatment recommendations, real-time tracking, and finally visualizing results. The innovation lies in creating a comprehensive healthcare ecosystem framework and an application that not only aids in COVID-19 diagnosis but also addresses broader health challenges. The main objective is to introduce a novel framework designed to simplify the development and construction of applications by standardizing essential components required for applications focused on addressing diseases. CronaSona includes two parts, which are stakeholders and shared components, and four subsystems: (1) the management information subsystem, (2) the expert subsystem, (3) the COVID-19 detection and forecasting subsystem, and (4) the mobile tracker subsystem. In the proposed framework, a CronaSona app. was built to try to put the virus under control. It is a reactive mobile application for all users, especially COVID-19 patients and doctors. It aims to provide a reliable diagnostic tool for COVID-19 using deep learning techniques, accelerating diagnosis and referral processes, and focuses on forecasting the transmission of COVID-19. It also includes a mobile tracker subsystem for monitoring potential carriers and minimizing the virus spread. It was built to compete with other applications and to help people face the COVID-19 virus. Upon receiving the proposed framework, an application was developed to validate and test the framework’s functionalities. The main aim of the developed application, CronaSona app., is to develop and test a reliable diagnostic tool using deep learning techniques to avoid increasing the spread of the disease as much as possible and to accelerate the diagnosis and referral of patients by detecting COVID-19 features from their chest X-ray images. By using CronaSona, human health is saved and stress is reduced by knowing everything about the virus. It performs with the highest accuracy, F1-score, and precision, with consecutive values of 97%, 97.6%, and 96.6%. Graphical Abstract


Introduction
The world has put the COVID-19 epidemic at the forefront as it spreads fast and terrifyingly all over the world.The virus infected around 698,777,806 people, with 6,947,449 recovered and 668,915,065 deaths1 [1].Addressing the epidemic requires the development of strategies to minimize infections.These strategies encompass the management and monitoring of existing cases, along with predicting the potential outcomes for individuals-be it recovery, death, or the emergence of new cases in the near future.This comprehensive approach involves collaboration across medical, engineering, and software domains.
Numerous systems, programs, and well-known websites dedicated to COVID-19 have surfaced, all with the shared objective of aiding people globally in navigating this perilous pandemic.These diverse platforms range from daily trackers, documenting the number of infections, recoveries, and fatalities worldwide, to others offering medical consultations to provide advice for individuals in quarantine.
The proposed framework, CronaSona, innovates by integrating diverse functionalities in one platform, providing reliable data, and fostering collaboration between stakeholders for a more robust healthcare infrastructure.
CronaSona app is a specialized mobile application for helping people get information about confirmed, recovered, and deceased cases.It can forecast the near future, trying to help people reduce the feeling of danger as it makes the virus statistically aware of its dimensions.It tracks patients by displaying infected people's locations using the Google Map location tracking system and also tracks COVID-19 cases by collecting data and then generating a forecasting model to forecast near-future death, recovery, and new cases.
Through the utilization of CronaSona, the uncertainty surrounding COVID-19 no longer instills fear, as users can benefit in various ways.Firstly, users can access medical guidance within their homes, learning essential safety measures and receiving assistance.Secondly, users possess a chest X-ray checker on their devices, eliminating the need to visit hospitals and mitigating potential risks.Thirdly, users can consult CronaSona for information on available hospital beds across their country.Fourthly, users can track areas with reported cases, identifying suitable hospitals and their locations.Finally, doctors can share advice through posts, with users engaging in discussions and seeking guidance.
The framework and application are designed for scalability, future disease detection, and global use.Continuous improvement is outlined for increased stakeholder roles, advanced detection models, and generalization to detect other diseases, making CronaSona a potential global health application.
In this paper, Section 2 provides an overview and summarization of related work conducted by various authors.Section 3 introduces Cronasona, the proposed framework for COVID-19 detection and forecasting.The examination of the proposed framework is detailed in Section 4, where an intelligent mobile application, CronaSona app, is developed.Section 5 discusses the results, followed by conclusions in Section 6, and prospects for future work are outlined in Section 7.

Related work
Many artificial intelligence methods and technologies are being used to detect diseases, especially in recent years, the detection of the emerging coronavirus (COVID-19).
Banja et al. [1] introduced a prototype utilizing advanced intelligent technology, aiming to provide a COVID-19 diagnosis and management solution to safeguard the well-being of individuals residing in the Kingdom of Saudi Arabia.The introduced expert system paradigms offer monitoring and control capabilities for COVID-19 patients, providing guidance to clinicians and patients using medical expertise while simultaneously alleviating the strain on healthcare providers and reducing healthcare expenses.
Asif et al. [4] proposed an optimized, lightweight shallow convolutional neural network (CNN) structure that effectively classifies patients' X-ray images with a minimal falsenegative rate.Our dataset consisted of 2541 chest X-rays.Through our model, they achieved an impressive accuracy rate of 99.68% in differentiating between healthy individuals and those infected with COVID-19.Additionally, our approach yielded excellent results in terms of precision, sensitivity, and specificity.Specifically, our proposed method demonstrated a precision of 99.66%, sensitivity of 99.66%, and specificity of 99.70%.
Oyelade et al. [5] presented CovFrameNet, a framework designed for the identification and categorization of COVID-19 presence in chest X-rays and CT images.This framework incorporates a sequential image pre-processing technique and employs a deep learning model for feature extraction, classification, and performance evaluation.The outcomes demonstrated that the proposed model achieved an accuracy rate of 0.1, recall/precision of 0.85, F-measure of 0.9, and specificity of 1.0.These results suggest that a CNNbased approach with image pre-processing capabilities holds promise for the initial screening of suspected COVID-19 cases, complementing the confirmation of COVID-19 detections based on RT-PCR methods.
Chen et al. [6] proposed a deep learning framework for the diagnosis and severity assessment of COVID-19 using chest CT scans.The framework comprises two modules: (1) the segmentation module, which extracts regions of interest and calculates the opacity percentage, and (2) the diagnosis module, responsible for identifying suspect cases and categorizing them into three groups: healthy, early stage, and progressive stage.To train and test the framework, a dataset of 150 CT exams was utilized.The obtained results include an impressive F1 score of 95.44% for COVID-19 detection, a robust F1 score of 90.87% for severity assessment, and an overall accuracy of 97.42%.
Motwani et al. [7] presented an improved framework for the automated identification and classification of COVID-19 patients using CT images.This framework utilizes a dense CNN model, transfer learning techniques, and a novel cross-entropy loss function to achieve efficient patient categorization.The proposed model has the potential to serve as a valuable clinical tool for diagnosing patients even before conducting pathogenic tests, thereby saving lives and preventing the spread of the virus.Furthermore, this model can be integrated into existing healthcare tools to accurately classify other pulmonary and chronic diseases.One significant advantage of this approach is its ability to expedite the diagnosis and treatment of COVID-19 cases.The proposed model demonstrates a prediction accuracy of 93.78% with a remarkably low false-negative rate of only 6.5%, indicating its effectiveness in diagnosing COVID-19-positive cases.
Kallel et al. [8] developed a tailored framework for monitoring and predicting the progression of COVID-19 that integrates various technologies such as machine learning (ML), cloud computing, fog computing, and the Internet of Things (IoT).This framework facilitated the collection and preprocessing of data, the implementation of classification models, and the training process using a combination of federated batch and stream ML services.Data were gathered from both medical devices (e.g., X-ray machines, lung ultrasound machines) and non-medical devices (e.g., bracelets, smartwatches) through streaming data sources.The proposed framework has the capability to incrementally learn and train from both real-time streaming data and historical stored data over time.
Pandianchery et al. [9] introduced the long short-term memory (LSTM) model for accurately predicting the number of active COVID-19 cases in each province of India.Our proposed model was trained on the data from one specific state, Maharashtra, and then tested on the data from the remaining provinces across India.The primary objective was to forecast the active cases in India from December 16, 2021, to March 5, 2022.Additionally, they experimented with various deep learning architectures, including Sim-pleRNN, LSTM, GRU, stacked RNN, stacked LSTM, and stacked GRU.These architectures were trained and tested on data specifically from Maharashtra.To ensure a comprehensive analysis, they conducted a detailed comparison of these architectures, considering different input window sizes and varying numbers of hidden units.
Hassan et al. [10] proposed a robust technique aimed at analyzing, predicting, and detecting COVID-19.This technique encompasses two primary tasks: infection forecasting and COVID detection.To achieve COVID detection, they employed two features: the scale invariant feature transform (SIFT) and the histogram of oriented gradients (HOG).In order to accurately predict COVID infections, they proposed a Gaussian model that facilitates reliable forecasts.For the purpose of COVID-19 detection, their technique leverages the histogram of oriented gradients and scale-invariant feature transform features.Additionally, they developed a CNN-based architecture called COVIDDetectorNet, which serves as an effective classification tool for detecting patients with COVID-19 and patients experiencing lung infections.Through our method, they achieved notable accuracy rates of 96.51% for two-class classification, 92.62% for three-class classification, and 86.53% for four-class classification.
Alruwaili et al. [11] developed a system to enhance the early diagnosis of the COVID-19 epidemic by accurately diagnosing chest X-ray images.The aim is to reduce the time required for diagnosing COVID-19 cases.Additionally, an ensemble of deep learning (DL) models is being constructed within a new framework.Furthermore, an improved DL model is being proposed, which incorporates a novel transfer learning algorithm to address the issue of overfitting, thereby making it more effective in real-time scenarios.
Ahmed and Eassa [13] have introduced a novel IoT and cloud-based blockchain model for the control of COVID-19 infection spread.This proposed model integrates IoT outdoor and indoor tracking technologies with a cloud computing-based blockchain system to effectively manage and minimize the spread of COVID-19.
Kanimozhi and Pradeep [14] proposed a framework that integrates preprocessing, feature extraction, and classification processes for the analysis of electroencephalogram (EEG) data.The preprocessing stage involves the application of principal component analysis (PCA) to eliminate artifacts from raw EEG data.Subsequently, independent component analysis (ICA) is employed for feature extraction, aiming to separate relevant features from the denoised data.The final step involves utilizing radial basis function neural network (RBFNN) for classifying the conditions based on the extracted features from the data.This comprehensive framework aims to enhance the accuracy and efficiency of EEG data analysis by systematically addressing preprocessing, feature extraction, and classification aspects.
Pradeep et al. [15] introduced a mathematical framework for modeling the rate of spread of the COVID-19 virus, utilizing an SIR-based epidemic model.Furthermore, an optimization control mechanism has been integrated into both susceptible and infected populations to regulate the spreading rate and enable early-stage diagnosis.
Table 1 between CronaSona and some of the previous mentioned frameworks/Models.
The previous works encountered several challenges and limitations in one or more of the following: -The absence of a collaborative approach among academic, research, healthcare, and governmental institutions may have limited the overall impact of previous works.-Certain frameworks and models demonstrated shortcomings in achieving high diagnostic accuracy and performance when it comes to detecting COVID-19 cases from chest X-ray images, particularly in the early stages of infection.-Some frameworks may not have employed robust forecasting models, affecting their ability to predict COVID-19 spread accurately.
-Certain frameworks may have faced challenges in adopting and integrating emerging technologies, such as deep learning models and forecasting tools.-The integration of different subsystems and components, as well as interoperability with existing healthcare systems, could be a challenge in some frameworks.-Previous works missed the communication and interaction mechanism among administrators, healthcare providers, and users could hinder the overall functionality of the system.-The level of user engagement and interaction in certain frameworks may have been insufficient.-Many existing frameworks and applications lacked scalability, making it difficult to handle a large volume of users and data.-Some frameworks were designed for specific regions or populations, limiting their generalizability to diverse demographic groups and global applicability.-Some frameworks may not have covered the entire healthcare ecosystem, leaving gaps in areas such as mobile tracking, expert advice, and comprehensive management information.

Materials and methods
This section covers the conventional procedure for the proposed framework, CronaSona.The strategy involves initial data preprocessing, followed by the utilization of a classification model derived from the convolutional neural network (CNN) model.The dataset utilized in this study is described.The proposed approach is initially depicted in Fig. 1.The framework parts are (1) stakeholder's part and (2) shared components part, while its subsystems are (1) management information, (2) COVID-19 detection and prediction, (3) mobile tracker, and (4) expert subsystem.

A. Stakeholder part
The stakeholders' part includes all types of system users.It encompasses a diverse array of system users, each playing distinct roles within the system.These stakeholders include administrators, users (comprising doctors, patients, and healthy individuals), and subscribed hospitals.These stakeholders collectively form a network that supports efficient healthcare management, diagnosis, and communication within the CronaSona ecosystem.The system facilitates seamless collaboration among diverse user roles, contributing to effective pandemic response and healthcare delivery.Table 2 explores the roles and responsibilities of each stakeholder category.

B. Shared component part
Shared components collectively form the backbone of CronaSona, ensuring that data is appropriately managed, shared, and utilized across different parts of the healthcare ecosystem.It plays a crucial role in facilitating seamless interactions and data management among various subsystems.The integration of these components enhances the overall user experience and contributes to the successful implementation of the proposed framework.It includes four components, which are (1) a knowledgebase that was collected mainly from two resources: the expert subsystem and the COVID-19 detection and forecasting subsystem.It contains knowledge about doctors' instructions, classified chest X-ray images, and rules discovered from the previous patients' electronic health records.(2) Database, which serves as a repository for stakeholder-related information, encompassing data such as user profiles, personal details, patient records, medication lists, hospitalization details, contact lists of patients, doctors' interactions with patients, treatment plans, and management protocols.(3) Electronic health record, which includes patients' reports and/or their chest X-ray.Stakeholders will communicate and/or interact with the framework subsystems through (4) the user interface.Table 3 summarizes the shared components part.

A. Management information subsystem
The management information subsystem within the CronaSona framework plays a vital role in collecting, processing, and presenting data gathered from various sources.It collects data from a range of sources, processes it, and presents it in a readable format for decision-making.It is responsible for managing information related to stakeholders.The main role is to manage collected data and disseminate information to other subsystems.In the data collection phase, data is collected from stakeholders and/or other subsystems.The collected data includes users' personal data, laboratory tests, and X-ray images provided by patients, supportive care, and treatment recommendations provided by doctors, hospitals' details and drug lists entered by admins, image classification and forecasting results provided by the COVID-19 detection and forecasting subsystem, and nearby carriers provided by the mobile tracker subsystem.According to the nature of the collected data, it is stored in a knowledge base, database, or electronic health record on the shared component's part.The processing phase includes rules and procedures associated with coordination between users, control, storing data, querying information, and/or analyzing results.Users have the ability to make inquiries or retrieve information from both the database and the knowledgebase.The data visualization phase includes knowledge representation, reports and analysis of results, treatment recommendations from doctors, drug lists for patients, or maps to represent nearby carriers.

B. Expert subsystem
The expert subsystem in CronaSona incorporates a life cycle approach to knowledge base development, including knowledge acquisition, validation, and verification.This systematic approach enhances the reliability of expert advice provided within the application.The development of the knowledge base goes through a life cycle comprising three phases: (1) knowledge acquisition, (2) knowledge verification and validation, and (3) the implementation of the inference engine.The knowledge can be collected either from expertise or from the detection and forecasting subsystem.After the information is collected, it needs to be revised by a knowledge engineer.The expert subsystem includes the following components: • An expert is a person who has special skills and/or knowledge in the medical field, such as doctors.• Inference engine: The inference is made on the knowledge base by applying logical rules to figure out new facts and/or information.

C. COVID-19 detection and forecasting subsystem
This subsystem is responsible for detecting COVID-19 carriers from their chest X-rays and forecasting the near future-the next 7 days-death, recovery, and new cases.It includes the following two phases: • The first phase, COVID-19 detection According to this phase, a ML classification model is used to classify users' chest X-rays.The classification result will be either normal or infected, depending on the infection percentage.Firstly, a user will input a chest X-ray image to the subsystem.Secondly, DL classification models will be used to process and classify this image in order to detect the degree of virus infection.Lastly, according to the classification model, the result will be displayed.It helps the patient get the closest possible result by displaying the percentage of infection with COVID-19.

Algorithm 1 Detection Module
The subsystem of the proposed model is depicted in Fig. 2 encompasses a sequence of six steps [16].These steps entail (1) collecting the dataset, (2) implementing data augmentation, (3) conducting data preprocessing, (4) establishing training and testing sets, (5) training and optimization, and (6) evaluations for assessing the overall performance.

Dataset collection and description
The sample of chest X-ray images covered two main classes: COVID-19 for positive cases and negative for healthy cases.The chest X-ray images are obtained from the COVID-19 Radiography Dataset. 2 It contains a total of 5863 X-ray images (JPEG), which are distributed into three folders: 624 images for testing, 5216 images for training, and 16 images for validation, each with two categories: pneumonia for COVID-19 images and normal for healthy images, as shown in Table 4.
As seen from Table 4, the distribution of images in the dataset is not suitable because it lacks balance among the classes, particularly in the test, training, and validation sets.For instance, in a medical imaging dataset with two classes (normal and pneumonia), having a highly imbalanced distribution can lead to biased model training and evaluation.In this case, there are substantially more pneumonia images than normal images in the training set, which may cause the model to be biased towards predicting pneumonia, affecting its generalization to new, unseen data.
To address this issue, data augmentation techniques are suggested.Data augmentation involves creating additional training samples by applying various transformations to the existing images.This helps in diversifying the dataset and improving the model's ability to generalize to different scenarios.The proposed approach mentions the use of multiple data augmentation techniques, such as flipping and rotation, to accurately capture and comprehend the various transformations that occur during the training phase of deep learning models.

Data augmentation
The application of data augmentation is a commonly employed method to increase the variety and volume of annotated training datasets by generating additional training samples through applying various transformations.The aforementioned objective is accomplished by the implementation of input transformations that preserve the corresponding output labels.The objective of this approach is to enhance the overall performance and outcomes of machine learning models by increasing the quantity of accessible data.Furthermore, varieties of photo augmentation techniques are utilized to enhance images within the realm of deep learning.The proposed approach integrates multiple data augmentation techniques, such as flipping and rotation, in order to accurately capture and comprehend the various transformations that occur during the training phase of the DL Models.Table 5 shows a comparison between the used data augmentation techniques, Flipping and Rotation.

Data preprocessing
The COVID-19 detection model being considered includes a preprocessing stage that is largely focused on enhancing the quality of the input image.The implementation of the preprocessing stage aims to address undesirable distortions and improve specific qualities present in the image.The procedure of image enhancement entails mitigating the adverse impacts of blurriness or noise present in the input image.The preprocessing stage involves the utilization of two approaches, specifically image scaling using the Nearest-Neighbor Interpolation technique, which is fast and computationally less intensive, and median filtering using the standard median dilter technique, which is effective for salt-and-pepper noise reduction.The process of image scaling is predominantly efficient in the resizing of input chest X-ray pictures.It is crucial in this operation to ensure that the quality of the input image remains unaffected by any form of degradation.The utilization of median filtering is employed to generate an image that is devoid of noise.The procedure entails the utilization of a non-linear filter, wherein the median value is employed to maintain the clarity of the image, particularly at its edges.The median value restoration has been accomplished by use gray level ranking instead of the noise value.
There are several libraries and modules that can be used to read and preprocess JPEG images, especially when working with Python and TensorFlow as a deep learning framework.For example, "tensorflow.image.resize" is used to resize images and "tensorflow.image.convert_image_dtype" is used to convert the image data type to a suitable format to be used by various DL models.Table 6 shows a comparison between image scaling and median filtering.

Establishing training and testing sets.
In the preliminary stage, data augmentation was utilized to tackle the challenge of imbalanced sample distribution in the raw X-ray images.After the augmentation process,  the improved dataset is then partitioned into three separate subsets, namely the train, test, and validation sets.

Training and optimization.
During the training and optimization phase, the Cro-naSona network model has been employed and requires fine-tuning to accurately identify the COVID-19 classes as either positive or negative.Hence, the model is optimized to achieve the most optimal decision-making model.The utilization of a validation error has been employed to enhance the optimization process.Following this, the optimal model has been acquired and subsequently employed in the decision-making procedure using the test dataset.The optimal network model achieved successfully classifies the different classes of COVID-19 and accurately evaluates the performance of the classification.

Evaluations.
The assessment of the predictive capabilities and effectiveness of the provided prediction models entails the application of five metrics.The evaluation of each model's performance involves the employment of many measures, such as accuracy, F-measure, recall, and precision [12].The determination of accuracy is achieved through the utilization of the following equation.where TPos is true positive, Tneg is true negative, Fpos is false positive, and Fneg is false negative.Precision is calculated using the following equation: Recall is computed using the following: F-measure is calculated using: (1) Accuracy = TPos + TNeg TPos + FPos + FNeg + TNeg This phase is divided into two stages.According to the first stage, data preparation, the subsystem will (1) read data from its source, the World Health Organization, as a JSON file; (2) extract data of interest; and (3) transform data to JSON format.In the second stage, model building and training, (1) the result accuracy is needed to be determined.(2) The forecasting model will be built.( 3) The model will be trained.( 4) The R2 error will determine whether to repeat this stage or not.An integrated mobile tracker subsystem is implemented to monitor individuals who may be carriers of COVID-19 by tracking their locations on a map.This system assists healthy individuals by providing the ability to check if COVID-19 carriers have been present at specific locations, thus enabling proactive measures.By displaying the locations of these carriers, the system aids in isolating them from healthy individuals and minimizing the potential spread of the virus.
The integrated subsystem utilizes Google Maps and a phone number list of confirmed COVID-19 cases to locate individuals, allowing for the identification of places they have recently visited.This information is then used to notify those who may have been exposed to the virus at these specific locations.By displaying a Google map, it becomes possible to highlight areas where suspected cases have occurred, thereby enabling healthy individuals to make informed decisions about the places they choose to visit.
Table 7 compares and highlights the key aspects of the management information subsystem, expert subsystem, COVID-19 detection and forecasting subsystem, and mobile tracker subsystem within the CronaSona framework:

Results and discussion
The subsequent section showcases the simulation outcomes of the proposed ideal model for forecasting the transmission of COVID-19.The implementation of the machine learning models was performed using the Jupyter Notebook version 6.4.6.The use of Jupyter Notebook enables the streamlining of Python code writing and execution, hence augmenting productivity and convenience.The assessment of the predictive accuracy is carried out by comparing the results obtained from the proposed model with those generated by other state-of-the-art deep learning algorithms.In addition, the assessment of the effectiveness of the proposed methodology frequently entails the application of characteristic curves and precision measures, commonly utilized to measure the performance of machine learning models.These metrics provide quantitative measures of the model's efficacy in accurately classifying instances, identifying true positives, true negatives, false positives, and false negatives, and evaluating the overall performance of the model.The accuracy metric measures the proportion of correctly classified cases relative to the total number of instances.Sensitivity, also known as recall or true positive rate, is the proportion of true positives accurately identified by the model.The accuracy findings obtained from the experimental investigation are presented in Table 5.Among all of the experimental models shown in Table 8, CronaSona model shows superior accuracy compared to other models.Specifically, its accuracy, F1-Score, Recall, and precision are 97%, 97.6%, 89% and 96.6%, respectively as shown in Fig. 3. Figure 4 shows the selected 6 DL classification models accuracy vs. epoch and loos vs. epoch.
The following Fig. 5 shows the confusion matrix of the 6 DL classification models VS.CronaSona.

Applying cronaSona
CronaSona was built to test the proposed framework functionalities.The aim of this application is to develop and test a reliable diagnostic tool using deep learning techniques to detect COVID-19 features from chest X-ray images.CronaSona would accelerate the diagnosis and referral of patients.This particular application also has the potential to be scaled up and used for more generalized, high-impact applications in biomedical imaging.Table 9 shows the main CronaSona functionalities.
Flutter is used as a cross-platform to develop Crona-Sona.In Flutter, the Scaffold class offers a range of widgets and APIs such as Drawer, SnackBar, BottomNaviga-tionBar, FloatingActionButton, and AppBar.It serves as a foundational component that facilitates the implementation of the application's basic material design layout.SQLite is used as a tool to build the CronaSona database.In order to build the mobile tracker subsystem, Google Maps is used through APIs.Tensorflow is used to run the training process for the classification models.Keras is chosen as an open-source, high-level deep neural network library to run on TensorFlow.Fbprophet is chosen as an open-source prophet for the forecasting process.It is a forecasting tool that Facebook developed and is available in both Python and R. The forecasting process is done based on data given from a JSON file, such as the following: The following code is used to training the dataset with performance accuracy 0.97.The following code is used for forecasting the next week day by day.
The following Fig. 6 shows the forecasting curve for the next 7 days.
The following Fig. 7 shows the CronaSona interface design.

Conclusion
In this paper, a healthcare ecosystem framework was introduced to help people face the dangerous COVID-19 epidemic.The proposed framework was designed for COVID-19 diagnosis and management.It consists of two parts: stakeholders and shared components, including four subsystems: the management information subsystem, expert subsystem, COVID-19 detection and forecasting subsystem, and mobile tracker subsystem.The push to develop mobile applications and eHealth solutions in response to the COVID-19 pandemic has resulted in a diverse range of apps and digital healthcare solutions being created.CronaSona was built based on the proposed framework as a step towards helping Egyptians to understand the prevalence and predict situations in Egypt, helping them to communicate with doctors, spread awareness, raise, and improve their psychological state by spreading the answers and trying to help people to each other in emergency situations.
CronaSona intends to contribute to managing the COVID-19 pandemic by providing a tool for diagnosis, forecasting, and tracking.It offers functionalities for stakeholders like administrators, hospitals, doctors, and patients.It helps the country serve people by directing them to the nearest specialized places available for their sites.This project also gives insight into how the country is performing in terms of curbing the spread and increasing citizen awareness.The accuracy findings show promising results compared to other models.It employs Flutter for cross-platform development, SQLite for the database, Google Maps API for the mobile tracker subsystem, Tensor-Flow for training classification models, Keras for deep neural networks, and Fbprophet for forecasting based on data from a JSON file.
The CronaSona framework, while promising, has certain limitations that need consideration.These limitations highlight areas for refinement and expansion, ensuring CronaSona's effectiveness, inclusivity, and adaptability to a wider range of healthcare scenarios beyond the current pandemic.The following points summarize CronaSona Limitations.:The implementation of machine learning models, including deep learning algorithms, is a key strength.The paper highlights the performance metrics of the proposed model, demonstrating its accuracy, F1-score, recall, and precision.The model, known as CronaSona, demonstrates superior accuracy (97%) compared to other state-of-the-art deep learning algorithms, as indicated in the presented results.This transparency in reporting results contributes to the credibility of the framework.11.Future-oriented approach: The proposed method outlines a roadmap for future work, emphasizing continuous improvement and addressing potential limitations.The focus on testing, stakeholder expansion, subsystem enhancements, and generalization for other diseases demonstrates a forward-looking and adaptive approach.12. Application in a specific context: CronaSona is tailored for the Egyptian context, addressing the specific needs and challenges faced by the population.This localized approach increases the relevance and effectiveness of the application in the given region.
The proposed CronaSona framework exhibits several limitations, which are:

Future work
To create comprehensive apps with numerous useful features consolidated in one location and supported by trustworthy data and frequent updates, more advancements are required.Strengthening the partnerships between academic and research institutions, healthcare providers, and government entities is essential to developing apps that enhance the quality of public information and bridge the gaps in healthcare exposed by the ongoing pandemic.This scoping assessment highlights the current domains in which COVID-19 apps are being developed and identifies areas for improvement and innovation.Collaborative efforts among key stakeholders will not only provide vital platforms for addressing the present crisis but also establish the groundwork for addressing future health challenges by building robust infrastructure.For that, several improvements might be made to the existing framework and the implemented application.CronaSona has to be subjected to a range of tests and experiments.Stakeholders need to be increased to cover different roles, such as nursing and/or adding a caregiver for patients.The management information subsystem needs to be improved by allowing users to register beds in hospitals.The detection and forecasting subsystem needs to be enhanced by training different DL models to improve performance based on different types of radiomics, such as CT scans and/ or MRI.The overall framework needs to be generalized to detect other diseases.In addition, CronaSona needs to be a global application with different languages to cover different countries.The following points stats several strategies and considerations that can be implemented: 1. Enhanced stakeholder roles: CronaSona wishes to consider expanding the roles of stakeholders to include additional healthcare professionals such as nurses and caregivers.It also needs to integrate features that allow caregivers to support patients and communicate with healthcare providers.2. Improved data quality and quantity: Ensure a diverse and comprehensive dataset, including high-quality chest X-ray images and collaborate with healthcare institutions and organizations to access a larger and more representative dataset.3. Multimodal approaches: Explore the integration of multiple modalities such as chest X-rays, CT scans, and clinical data for a more comprehensive understanding of COVID-19 cases and investigate the use of advanced imaging techniques to enhance diagnostic capabilities.4. Improvements in management information subsystem: CronaSona needs to enable users to register available beds in hospitals through the management information subsystem.It also needs to implement features that provide real-time updates on hospital capacity and resource availability.5. Enhanced detection and forecasting subsystem: Cro-naSona deep learning models need to be trained on diverse datasets, including different types of radionics such as CT scans and MRI, to improve performance across various medical imaging modalities.Additionally, it needs to implement continuous learning mechanisms to adapt the models based on emerging patterns in the data.6. Explainability and interpretability: Utilize interpretable machine learning models and techniques to enhance user trust and implement models that provide clear explanations for their predictions, ensuring transparency in decision-making processes.7. Globalization and multilingual support: The future vision for CronaSona includes global application by incorporating multilingual support, aiming to cater to diverse populations and regions and considers collaborating with international healthcare organizations to ensure the framework is adaptable to diverse healthcare systems.8. Usability and user interface enhancements: Conduct user testing to gather feedback on the user interface and overall usability then implement improvements based on user feedback to enhance the overall user experience.9. Collaboration with healthcare institutions: Strengthen partnerships with academic and research institutions, healthcare providers, and government entities to ensure the framework is aligned with the latest healthcare standards and guidelines.In addition, CronaSona needs to establish a mechanism for regular updates and collaboration to address evolving healthcare challenges.10.Comprehensive testing and experimentation: Conduct experiments to validate the effectiveness of the forecasting models and make adjustments based on the results.Additionally, focus on security audits and performance testing, to ensure robustness and data security.11.Generalization for other diseases: Explore the possibility of generalizing the framework to detect and manage other diseases, expanding its utility beyond COVID-19.12. Community engagement and awareness: Implement features that facilitate community engagement and awareness, encouraging users to actively participate in the collective effort to combat the pandemic, and incorporate educational resources to raise awareness about preventive measures and public health guidelines.13.Continuous improvement and updates: Establish a system for regular updates and continuous improvement based on emerging healthcare trends and user feedback.Stay informed about advancements in technology and healthcare practices to incorporate the latest innovations into the framework.

Fig. 4
Fig. 4 The applied five DL classification model accuracy and loss for each epoch.a Inception_V3 accuracy, b Inception_V3 loss, c Inception_ResNet50_V2 accuracy, d Inception_ResNet50_V2 loss,

4 .
Mobile tracker reliability: The effectiveness of the mobile tracker subsystem is contingent on the willingness of individuals to share their location data.Privacy concerns and data security issues might hinder the widespread adoption of this feature.5. Limited stakeholder roles: The framework's stakeholders are currently limited to administrators, hospitals, doctors, and patients.Including additional roles such

Table 1
A comparison between CronaSona and related frameworks

Table 2
Stakeholder roles and responsibilities

Table 3
The shared component parts

Table 4
The distribution of testing, training, and validation sets used in the dataset

Table 5
A comparison between data augmentation techniques (flipping and rotation)

Table 9
CronaSona functionalities JSON file for the confirmed, recovered, and dead cases Database -Includes saved data about stakeholders, and their communication posts and/or comments -Data about drug list Electronic health record -Includes patients' uploaded data, tests, and/or x-ray images