CellStrat AI Platform for Healthcare Applications
#CellStratAILab #disrupt4.0 #WeCreateAISuperstars
Recently, we conducted our first Product Showcase for our new world-class Healthcare AI application platform.
Given the times, COVID-19 detection is our top focus in terms of product development, but we also have other exciting healthcare AI applications being developed related to brain, eye and cancer issues.
The following demos were presented in our Healthcare Product Showcase :-
Overall Solution Architecture of CellStrat AI Platform :-
Niraj Kale provided an overview of the overall CellStrat AI Platform, which is hosted on the Amazon Web Services (AWS) in a SAAS model.
Our platform consists of AWS EC2 instance, S3 storage and lambda integration. The Cloud APIs for AI models are exposed via AWS Sagemaker.
AI models can be trained on AWS GPU instance or pre-trained models can be deployed.
Each model is deployed in a separate container and model endpoints are REST APIs.
The AWS APIs can be accessed by both CellStrat Web Frontend clients or programmatically by third-party systems.
COVID-19 Detection Chest X-Ray Analyzer :-
Our AI Researcher Bismillah Kani has created a innovative solution for detecting COVID-19 from Chest X-Rays.
This solution uses pre-trained RESNET model to extract image features from a large scale Chest X-Ray NIH dataset. There are three classes in this supervised dataset – Normal, COVID-19 and Pneumonia.
The COVID and pneumonia classes are under-represented. For this the model does data augmentation as well as oversampling of under-reported classes using a PyTorch API.
The model interpretation was also added for purposes of achieving explainability. This is done via the Grad-CAM algorithm, a technique to visually explain a decision/prediction made by CNN based models to make it more transparent and explainable. It is a way to understand what part of the images influenced the model to make a decision/prediction.
The model was deployed on AWS using S3 storage and Sagemaker.
Web Workflows and AWS Setup for Chest X-Ray solution :-
Bhavesh Laddagiri has developed an intuitive web interface for our Chest X-Ray solution and also set up Amazon Web Services (AWS) Sagemaker APIs for web-to-backend integration.
We used React frontend for web client development which consumes AWS Lambda APIs for model access.
COVID-19 Detection and Segmentation from CTSCAN images :-
Jani Basha demonstrated an interesting solutions on how to detect, localize and segment COVID-19 from CTSCAN images.
CT Scan or Computed Tomography Scans are 360 degree images of certain body parts.
We trained a COVID detection model using CT Scan images. It also included Image Segmentation of Lung regions and Lesion regions. Lung masks help zero in onto the lung regions and lesion masks help pinpoint the affected region in the lungs.
A DenseNet model was used for classification task and UNET based image segmentation model was used for segmentation task.
A more sophisticated INFNET architecture was also used for image segmentation.
Diabetic Retinopathy from Eye Scans :-
Amit Kumar demonstrated an intuitive solution for detecting early stages of Diabetic Retinopathy using Vision AI.
Diabetic retinopathy is a diabetes complication that cause vision loss for people with diabetes.
A more technical explanation is – It is a microvascular (damage to small blood vessels) complication of both type 1 and type 2 diabetes mellitus.
Early stages of diabetic retinopathy often don’t have symptoms.
Deep Learning AI algorithms (Vision) can analyse eye images to detect early signs of diabetic retinopathy such as haemorrhaging.
Brain Lesion Detection from MRI images :-
AI Researcher Niraj Kale demonstrated an interesting solution for detecting lesions or anomalies in the human brain from Brain MRI images.
A Brain Lesion describes destruction to any part of brain. It could be due to accident, or natural causes such as tumors of the brain.
The heterogeneous nature of TBI (traumatic brain injury) lesions poses challenges in using discriminative models. The healthy and non-healthy tissue can overlap easily making anomaly determination difficult.
Hence we detect brain lesions using a 3D CNN on MRI images, so as to analyze the brain tissue in 3D space. Here we operate on voxels (3D points) instead of 2D image pixels.
The model uses 3D CNNs, which convolve over 3 dimensions of the input.
To detect brain lesions, we use a multi-scale 3D CNN with two convolutional pathways. The Conditional Random Fields (CRF) help remove false positives.
The results are quite interesting.
CellStrat Advanced Product Internship Program :-
CellStrat AI Lab is a leading AI Lab and is working on the cutting-edge of Artificial Intelligence including latest algorithms in ML, DL, RL, Computer Vision, NLP, Graph Theory etc.
Interested in working in our world-class Healthcare AI Product team (as described above) ? Wish to gain FULL-STACK AI MODEL DEVELOPMENT AND DEPLOYMENT experience on our AWS Cloud Platform ? If yes, enroll in our Advanced Product Internship Program. More details and enrollment here : https://bit.ly/CS-CPI
Questions ? Please feel free to call me at +91-9742800566 !
Co-Founder & Chief Data Scientist, CellStrat