Explainable AI based Covid-19 Tele Health Solution


Convolutional neural network techniques have feature extractors within the defined approaches. Neural networks can help in feature extraction. However,  they cannot give explanations of how the extraction was done. For instance, it is easy to train a neural network to distinguish pictures of tigers from pictures of rabbits. It is tough to explain why a certain picture is a tiger. The algorithms  cannot elaborate what distinguishes tigers from rabbits.

However classical methods can explain in theory the differences in population genetics. Convolutional neural networks with Explainable AI is the most successful deep learning modeling technique for image processing. In genomics, In the next sections, you will see the details about the Omni Channel Telehealth platform, Deep Learning & Explainable AI models, Architectural components, Implementation and What’s Next.

 

OMNI Channel Telehealth Platform

First, let us look at how the omni channel tele-health platform is built.

We have built an Omni Channel Telehealth Platform powered by AI Deep Learning for doctor’s community and Telehealth support for patients during COVID-19 lockdown.   Doctors can extend Telehealth support using medical practice management solutions of the platform. Patients can order medicines online channels like desktop, web, mobile, voice, WhatsApp, and IVR through integrated pharmacies on the platform.

 

 

Figure 2: Omni Channel Telehealth Platform

Explainable AI based Deep Learning Platform has digital, voice, and knowledge assistants to provide information to the doctor using the knowledge base. AI techniques used in the platform are explain ability, inference engines, NLP/NLU, Conversational agents, Intelligent Assistants, Deep Learning, Knowledge studio, machine learning, knowledge mining, cognitive search, and decision-making applications.

This platform is based on polyglot microservices architecture using Java, Node.js, PHP, Angular, React, Laravel, Python, Apache SuperSet, MongoDB, MySQL, and Jenkins. Google dialog flow and Zoho IQ are used for voice assistants.

Rumours and fears were coming during the lockdown. The idea was to have authentic information that is fact-checked like  Whitecoats COVID-19 Campaign. The campaign had curated feeds for doctors which comes from authorized disease control organizations.

 

 

Figure 3: Knowledge base

We are enhancing the knowledge base which provides feeds to have learning capabilities. Self-manageable knowledge base has capability to maintain a disease registry and the registry can be updated from various popular data sources such as Orphanet, John Hopkins, NORD, CDC, and WHO. The current initiative will help in detection of the future upcoming undiagnosable disease which might turn into pandemic. The goal is to ensure early warnings and prevent the future dangerous diseases not to turn into pandemic.

Now, let us look at the deep learning & explainable AI modeling.

 

Deep LEARNING & Explainable AI

 

First, let us look at Deep Learning models and techniques used in the telehealth platform.

Deep learning is deep structured learning. It belongs to the category of machine learning AI based techniques. The deep learning model has multi-processing layers. The layers process the data which is the input and passes it through different layers which identify the specific patterns and representations. These layers classify the data based on the type of the input. Deep learning has been applied in various fields such as drug discovery, development, bioinformatics, plant genomics, toxicology and genome analysis. Computer vision, explainable AI and machine learning is also used for remote health monitoring and health care data analysis.

Now, let us look at the explainable AI model.

Explainable AI model captures the following while presenting the results:

  • What were the key factors in passed and failed scenarios?
  • What were the features extracted from the image and patient profile?
  • What were the disease specific rules used for checking the patient condition?
  • Which rules failed in the failed scenarios?
  • Which validation rules passed in the failed scenarios?

 

 

Figure 4: Explainable AI

Now, let us look at the implementation of the tele-health platform in hospitals and clinics.

Implementation

Telehealth platform is implemented in around 4000 clinics through mobile, WhatsApp, IVR and web applications for doctors, medical staff, and patients. An average of 5 patients per doctor are using the tele-health platform for consultations using video, audio, and chat channels. Doctors can dictate and prescribe by voice the patient notes and

medication. Doctor’s notes can be stored and retrieved on the platform.  Patients can order medicines using online e-pharmacy and use e-Pathology for lab diagnostic methods. The implementation plan succeeded in rolling out over 3000 doctor apps and websites under the WhiteCoats Practice Plus program, and over 30,000 unique doctor’s visits on a Monthly Average User basis to the WhiteCoats Network app and web content for

articles, cases, and interactions.

In the next section, let us look at what’s next in the platform roadmap.

WHAT’S NEXT

Enterprise AI has neural network techniques such as ANN, CNN, and RNN. Machine learning algorithms use neural network methods for data analysis and predictive analytics in healthcare. Machine Learning Models help in predicting the patient’s mortality rates and life expectancy.      Medical Expense Prediction Models can be developed using patient history & medical claim history for hospitals and clinics. Patient’s genetic data and health records will be helpful in precision medicine, disease analysis and drug discovery. Cure for Rare and undiagnosable diseases can be invented using the big data available related to historical patient genetic and health records. Knowledge base will help in early detection, prevention, and treatment for unknown pandemics.

AI Deep learning platform can be enriched with biometrics, multilanguage support, knowledge assistants, integrations with connected systems, and IOT.   Instead of software development, dynamic program creation will be the next enhancement to the platform.  An AI coder using neural sketch to change the technology stack will be the next generation AI Deep learning platform. No Code Automation platform which has the capability of the process design, prototype design, and business workflow design features will be the next step to the platform. Generating software and deployment of the code using DevOps are the features of the No Code Automation platform.

 

 

Figure 5: Next Generation Ideas


Recent Posts

View All Posts