About the Customer
The client is a global biopharmaceutical company focused on discovering, developing, and delivering innovative medicines for patients with serious diseases. Their medicines are helping millions of patients around the world in disease areas such as oncology, cardiovascular, immunoscience, fibrosis, and others.
Business Challenge
The Client wanted to reduce the tedious manual effort of clinicians reviewing X-rays and detecting and classifying diseases. The manual approach also led to increased false diagnoses and suboptimal outcomes. The client was using Python and R for such business cases and mostly had a manual approach. Using traditional methods of image classification was not yielding good results and manual efforts were very time-consuming and not cost-effective.
Brillio’s Solution
Brillio assessed their business and explored the features of AWS SageMaker. We also did a model comparison between AWS and traditional methods, aiming to migrate from traditional modeling to AWS using its cloud services.
The steps taken for analysis were as below:
- Pre-process the images to bring different image properties to uniformity
- Densenet 121 architecture was used for object detection
- Built a docker container with Amazon SageMaker to train & deploy our own algorithms
- Multiple models were built and trained to predict better accuracy
- Models with higher accuracy are deployed
The steps of the implementation and the approach:
- Developed the capability to accurately detect 15 diseases
- Built an ML algorithm to identify diseases from the data collected through X-ray reports
- Developed three candidate classification models using deep neural networks to train the algorithm.
- Built a hands-on lab to train the models and detect diseases in X-rays from multiple patients
- Scaled and deployed the AI solution to process more than 100K X-rays / day
- Technologies used: AWS Sagemaker, Tensor flow APIs
Business Impact
Following Brillio’s implementation, the client was able to:
- Detect anomalies in X-rays by using trained algorithms of AWS Sagemaker
- Generate post-disease detection reports for further analysis
- Develop three candidate classification models using deep neural networks to detect multiple diseases
- Process more than 100K X-rays/day by scaling and deploying the AI solution
- Accurately detect 15 diseases
- Optimize the time and reduce the cost of detection among 15 probable diseases
The results led to:
- 82-87% ACCURACY achieved in disease detection
- ~70% REDUCTION in cost to detect 15 diseases
- ~85% REDUCTION in time required to detect diseases
