Two researchers, Dr. Madhu Nair from the Artificial Intelligence & Computer Vision Lab, Cochin University of Science and Technology (CUSAT), and Dr. Asha Das from Union Christian College, India have developed a novel approach to help diagnose breast cancer in its early stages.
ADVERTISEMENT
(For insights on emerging themes at the intersection of technology, business and policy, subscribe to our tech newsletter Today’s Cache.)
Dr. Das and Dr. Nair have created a model called NAS-SGAN, which can distinguish the different cancer grades. Their model leverages deep learning and uses labelled and unlabelled images together to achieve high accuracy.
ADVERTISEMENT
They have worked with Intel on this project. For the best results, the deep learning solution which they used needed to process high-resolution images but was limited by the GPU’s inability to hold the entire AI model in memory. To help overcome this, Intel helped the researchers with a technology architecture based on its Xeon Scalable processors.
Dr. Madhu Nair, Cochin University of Science and Technology (CUSAT) said, “While we had worked out a solution consisting of discriminator and generator models trained in an adversarial manner using both labelled and unlabelled samples, we were worried whether it would work or not. We shared our problem with the Intel team and were extremely happy that they immediately understood the importance of this work. They gave us the opportunity to use this distributed architecture and helped us to improve the model and shared optimisations with us to get it working.”
NAS-GAN works in two phases: A GAN (Generative Adversarial Network) is used to create images that are indistinguishable from genuine histopathological images. The GAN is trained using unlabelled images, which are relatively easy to obtain.
ADVERTISEMENT
The new images are used to help the solution understand the data distribution. The GAN discriminator is then trained with the labelled images to predict the cancer grades. The NAS-SGAN approach uses four servers based on Intel Xeon Scalable processors which train the solution in parallel, with 192GB of memory per server. Intel Optimization for TensorFlow meanwhile makes it easy to use acceleration features in the processors.
Intel helped researchers with technology architecture to develop an AI-based model that detects breast cancer more accurately. | Photo Credit: Special Arrangement
Dr. Das and Dr. Nair compared the performance of NAS-SGAN with 10 other GAN algorithms used to detect breast cancer. The NAS-SGAN algorithm addresses the shortcomings of other GAN models for breast cancer screening by adding the ability to grade cancer images.
NAS-SGAN achieved an accuracy of 98%, approximately 10% higher than the next-rated GAN (WGAN-GP). Precision was 97%, 18% higher than WGAN-GP. NAS-SGAN achieves high accuracy results even when using a limited amount of annotated data, which helps minimize the time-consuming and labour-intensive process of classifying images.
ADVERTISEMENT
Earlier, using other GANs, a physician had to study the histopathological image to grade it. The automated scoring of NAS-SGAN helps to streamline the diagnosis process and analysis process, helping to improve consistency and accuracy in grading.
Santhosh Viswanathan, Managing Director-Sales and Marketing Group, Intel India, said “We are overjoyed to have collaborated with Dr. Das and Dr. Nair to help fight this deadly disease, and we look forward to more such collaborations with the world’s brightest scientists.”
Going forward, the researchers are now looking at how a similar approach could be used for mortalities resulting from cerebral aneurysms and classifying polyps from endoscopies.
As per WHO, 2.3 million women diagnosed with breast cancer in 2020 resulting in 685,000 deaths.