Health

New deep learning algorithms can detect gene mutations in colorectal cancer more efficiently

Spatial map of colorectal cancer tissue sections created by the IDARS algorithm. Map surrogate measurements of instability (red) or stability (green) of DNA microsatellite in tumors. Tissue areas without overlays are non-tumor. Cases of colon cancer with high microsatellite instability are generally more likely to respond to expensive immunotherapeutic treatments.Credit: University of Warwick

A new deep learning algorithm created by researchers at Warwick University can detect the molecular pathways and development of major mutations that cause colorectal cancer more accurately than existing methods. This means patients can benefit from targeted therapies with faster turnaround times and lower costs.

State and key of molecular pathways involved in development for the rapid and efficient treatment of colorectal cancer mutation You need to determine the rate of cancer. Current methods for doing so include costly genetic testing, which can be a slow process.

However, researchers at the School of Computer Science at Warwick University are looking for ways to use machine learning to predict the molecular pathways of three major colorectal cancers and the status of hypermutated tumors. An important feature of this method is that you do not have to manually annotate the digitized images of cancerous tissue slides.

The paper published today 19th, “A weakly monitored deep learning framework for predicting the molecular pathways of colorectal cancer and the status of major mutations from routine histology.”NS In the October journal Lancet Digital Health, Researchers at the University of Warwick Machine learning From all slide images of colorectal cancer slides stained with hematoxylin and eosin, three important mutations can be detected as an alternative to the current study regime for these pathways and mutations.

Researchers propose new iterative draw-and-rank sampling algorithmYou can select representative sub-images or tiles from images of the entire slide, without the need for detailed cellular or regional annotation by a pathologist. In essence, new algorithms can harness the power of raw pixel data to predict clinically important mutations and pathways in colon cancer without human interception.

Iterative draw-and-rank sampling works by training a deep convolutional neural network to identify the image region that best predicts the major molecular parameters of colorectal cancer. An important function of repetitive draw-and-rank sampling is to enable systematic and data-driven analysis of cell composition of image tiles that strongly predict colorectal molecular pathways.

The accuracy of repeated draw-and-rank sampling has also been analyzed by researchers, and their algorithms are significantly more accurate than currently published methods for predicting the molecular pathways and major mutations in the three major colorectal cancers. It was proved.

This means that patients may be stratified for targeted therapy using new algorithms compared to extensive post-validation sequencing or special staining-based approaches. ..

Dr. Mosinviral, lead author of the study and data scientist at the University of Warwick’s Center for Tissue Image Analysis (TIA), said:Detection Molecular pathway Major mutations with Colorectal cancer Then select patients who are more likely to benefit from targeted therapy at a lower cost with faster turnaround times. We are also looking forward to the important next step of validating algorithms in a large multicentric cohort. “

Professor Nasir Rajpoot, director of the TIA Center in Warwick and senior author of the study, commented: cancer.. The main advantage of the iterative draw-and-rank sampling algorithm is that it does not require time-consuming and labor-intensive annotation by a professional pathologist. These findings open up the potential for repeated draw-and-rank use. sampling Select patients who are likely to benefit from targeted therapy compared to sequencing or special marker-based approaches, and do so at a lower cost and with faster turnaround time.

“We are now aiming to conduct a large-scale multicenter validation of this algorithm, paving the way for its clinical adoption.”


Use machine learning to find mutations in similar genomic sequences in cancer samples


For more information:
A weakly monitored deep learning framework for predicting the molecular pathways of colorectal cancer and the status of major mutations from routine histology, Lancet Digital Health, DOI: 10.1016 / S2589-7500 (21) 00180-1

Quote: New deep learning algorithm more efficient for colorectal cancer gene mutations obtained on October 19, 2021 from https://medicalxpress.com/news/2021-10-deep-algorithm-genetic-mutations-colorectal Can be detected (October 19, 2021). html

This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. The content is provided for informational purposes only.

Back to top button