An estimated 17.7 million people die annually across the world due to Cardiovascular diseases. The workload on radiologists increased by 1300% in the number of CT scans interpreted and 540% in the MR images.
Cardiovascular MRI is required by physicians and radiologists to understand the reasons behind a person’s illness and chalk out a relevant treatment process. Many tools that are available in the market can be used to view and analyze these MRI results. Some of the existing MRI analysis tools are CVI42, HARP, Sygnovia, etc.
Datasets and Augmentation
- Our data set consists of 25 case studies. Only 5 case studies were labeled.
- Due to the small size of the data set, we performed data augmentation as below using traditional algorithms
- Rotation,
- Cropping
- Image Transformation &
- Feature extraction methods.
- Rotation,
- We used the latest algorithms and Random forest trees for labeling unlabeled data set.
- Deep neural networks work better on large data sets for better Network Generalization.
Our Method
In Healthcare, finding large labeled data sets is a challenge. The biggest advantage of our Feature Extraction method is that our algorithms perform extremely well on small data sets. Other learning algorithms require a lot of labeled data sets.
Method – Deep Learning Architecture (LVRVNet)
Tech Vedika is currently developing a Cardiovascular MRI Analysis Tool which is powered by Artificial Intelligence. We are currently working on the full ventricular function assessment. The software consumes the entire patient data and separates out the short axis DICOMS. It is capable of fully automatic Deep Learning and Active learning-based contour detection, which calculates the ventricular functional parameters.
Developed a Fully Automated Segmentation method for MRI images using a deep learning model.
Calculated Values
Stroke volume, End Systolic Volume, End Diastolic Volume, Ejection Fraction, Heart Rate, Cardiac Index, Cardio Output, Body Surface Area.
About the Author
