Artificial intelligence (AI) machine learning is emerging as a useful tool to assist cancer clinicians. A new study published in Natural Medicine shows how combining AI edge computing with swarm learning (SL) can detect cancer from patient data while preserving patient privacy and information governance.
A team of researchers affiliated with the University of Leeds in the UK, and various institutions and hospitals in Germany set out to discover if swarm learning can be applied to AI machine learning in order to predict molecular changes directly from histology images. Histology, also known as microanatomy, refers to the study of the microscopic anatomy of biological tissues.
“Digitized histopathology images contain a wealth of clinically relevant information that AI can extract,” the researchers wrote. “For example, deep convolutional neural networks have been used to predict molecular alterations of cancer directly from routine pathology slides.”
The researchers point out that in 2018, a different team of scientists demonstrated proof of this concept for non-small cell lung cancer using deep learning, and subsequently, “dozens of studies have extended and validated these findings to colorectal cancer (CRC), gastric cancer, bladder cancer, breast cancer, and other tumor types.”
In data science, swarm learning refers to a decentralized data privacy-preserving framework that leverages blockchain for AI machine learning. In swarm learning, the training of the AI algorithm happens locally at the edge rather than via a centralized server. Separate computer systems can jointly train a machine learning algorithm using swarm learning without compromising data privacy.
For this study, the researchers used Hewlett Packard Enterprise (HPE) implementation of Swarm Learning and open-sourced their source codes. Data from three large repositories stored in three separate computing servers were used for training the algorithm.
“We trained AI models on three patient cohorts from Northern Ireland, Germany, and the United States, and validated the prediction performance in two independent datasets from the United Kingdom,” the scientists wrote. “Our data show that SL-trained AI models outperform most locally trained models and perform on par with models that are trained on the merged datasets.”
The scientists showed that swarm learning can enable AI machine learning predictions of clinical biomarkers in solid tumors, especially for two critical biomarkers for colorectal cancer—BRAF and MSI.
With this proof-of-concept, this technique is not limited to cancer prediction; it can be used for analyzing images for other diseases and disorders.
“In the future, our approach could be applied to other image classification tasks in computational pathology,” wrote the researchers.
Copyright © 2022 Cami Rosso All rights reserved.