New AI Platform Increases Speed and Accuracy Diagnosing Pediatric Cancer

"As the burden of cancer increases worldwide, the complexity of cancer diagnostics is expected to grow unless new methods are developed. Our platform can be used at any hospital to increase the speed and accuracy of diagnosing cancer, even for rare types."

Dr. Adam Shlien, Senior Scientist, Genetics and Genome Biology, The Hospital For Sick Children

Scientists at The Hospital for Sick Children (SickKids) in Toronto are using AI to increase the speed and accuracy of diagnosing cancer. Dr. Adam Shlien and his colleagues have developed an AI based platform that can can classify every known type of childhood cancer and match a diagnosis for 85% of pediatric cancers. The new AI platform can help doctors and researchers identify specific cancer types faster and more accurately and can help researchers develop new therapeutics. The platform is already used in research at cancer centers around the world and Dr. Shlien believes that this platform has the potential to become a universal test for diagnosing pediatric cancer.

Dr. Adam Shlien is an Associate Director in the Department of Pediatric Laboratory Medicine and Senior Scientist in the Genetics & Genome Biology program. His current research focuses on analyzing pediatric cancer genomes using the most advanced sequencing tools. He's working to find genomic mutations that drive cancer development and understand how they alter the somatic transcriptome. His lab is introducing next-generation genomics technologies including genome sequencing into oncological practice and working to expand the eligibility for immunotherapy in the SickKids cancer sequencing program. A paper on their AI platform entitled Diagnostic classification of childhood cancer using multiscale transcriptomics was published this week in Nature Medicine.

Study Overview

Using neural networks to diagnose childhood tumors

1. Scientists designed a CNN called RACCOON (Resolution-Adaptive?Coarse-to-fine?Clusters?OptimizatiON) to be used as a diagnostic and prognostic aid for childhood tumors.
2. They also validated a classifier for childhood cancer called OTTER (Oncologic TranscripTome Expression Recognition) to classify new patients’ tumors.
3. They used a reference set of 2,178 child tumors and 9,400 adult tumors.
4. They analyzed 13,313 individual cancers and built an atlas of pediatric cancer using a novel machine-learning algorithm.
5. They performed an in-depth annotation of 162 clusters representing the major pediatric tumor families.
6. They defined childhood-specific cancer subtypes and investigated their internal differences in gene expression.
7. They used the atlas to explore age-associated changes
8. They noted changes in survival, age, sex and underlying genomic alterations
9. They noted key genes differentiating them from their adult counterparts.

"We were able to see, for the first time, subtle differences within cancer subtypes. Childhood cancers display more transcriptional variability—the number of the genes expressed in a cell—than adult cancers. This gives us a radically new way to look at cancer and potentially identify the prognosis of cancers, and the possibility of changing our understanding of cancer."

Dr. Adam Shlien, a Senior Scientist, Genetics & Genome Biology program, The Hospital For Sick Children

Study Highlights

1. The platform identified clusters for most major types of pediatric leukemia, brain tumors, and solid cancers.
2. The top three performing CNNs were integrated into an ensemble called OTTER which can output highly consistent predictions in just a few minutes.
3. The platform was able to match or clarify the diagnosis for 85% of childhood tumors in a prospective cohort.
4. The platform identified significant differences between cancer types and allows researchers to identify 455 subtypes of cancer.
5. Some tumors were divided into subgroups that predicted outcome better than current diagnostic approaches.
6. They found that tumors were sometimes unexpectedly grouped due to common lineages.
7. For nearly every recognized pathological classification of pediatric cancer, there was a corresponding transcriptional cluster.
8. For example, in brain cancers, one can differentiate subtypes of medulloblastoma, gliomas, as well as those with specific mutations.
9. Different histotypes were occasionally brought together into one cluster, indicating unexpected, shared expression programs or a common cell of origin. Within the hierarchy of brain tumors was a small but highly specific cluster of young childhood tumors. This data suggests that they are a distinct entity, independent of the location in which they arise.
10. Using this approach, they found 4 subtypes of neuroblastoma that have substantial differences in immune activity, differentiation level, and survival.?

Using Advanced AI In Cancer Sequencing

To diagnosis pediatric cancer, doctors rely on the microscopic examination and detection of specific proteins. The accuracy of current methods is variable, and improvements are not usually shared between hospitals. The Hospital for Sick Children has developed a strategic plan called SickKids 2025 which involves using AI and big data to provide state of the art precision medicine to children. They have developed a program called AI in Medicine for Kids that provides a platform and tools to facilitate implementation of AI into clinical practice. Through collaborations with leading researchers, computer scientists and clinicians, SickKids has established the infrastructure, policies, and processes needed to get advances in AI research to sick children sooner.

In December 2022, Dr. Shlien and his colleagues published an important study as part of the?SickKids Cancer Sequencing program. The study used next generation sequencing to deeply sequence over 860 cancer-associated genes. They also sequenced the complete genomes and transcriptomes, the complete set of both coding and non-coding RNAs of tissue, of 300 paediatric, adolescent and young adult patients with rare, relapsed and hard-to-cure cancers, using a novel machine learning algorithm.

I'm sharing some of Dr. Shlien's comments about hypertranscription from an article entitled "Hypertranscription by Tumors Is Linked to Poorer Cancer Outcomes: Study" published in The Scientist on December 13, 2022.

"Why hypertranscription is correlated with worse outcomes is still a mystery."

"Any hypertranscription was associated with poorer survival prognoses above and beyond stage, grade, and tumor subtype.”?

"Patients with tumors that had higher levels of hypertranscription did not survive as long as patients whose tumors had lower levels of hypertranscription."

"All of the genes are elevated above baseline level. It’s as if there’s a person with a megaphone who is amplifying this really corrupted message across the entire genome."

“The way that transcription factors drive hypertranscription is really through the absence of their suppression. . . . So instead of accelerating . . . you’re taking your foot off the brakes.”

“We think these patients (with higher levels of hypertranscription) should become eligible for immunotherapy based on the hypertranscription."

References

Diagnostic classification of childhood cancer using multiscale transcriptomics, Nature Medicine, March 17, 2023

Authors: Federico Comitani, Joshua O. Nash, Sarah Cohen-Gogo, Astra I. Chang, Timmy T. Wen, Anant Maheshwari, Bipasha Goyal, Earvin S. Tio, Kevin Tabatabaei, Chelsea Mayoh, Regis Zhao, Ben Ho, Ledia Brunga, John E. G. Lawrence, Petra Balogh, Adrienne M. Flanagan, Sarah Teichmann, Annie Huang, Vijay Ramaswamy, Johann Hitzler, Jonathan D. Wasserman, Rebecca A. Gladdy, Brendan C. Dickson, Uri Tabori, Mark J. Cowley, Sam Behjati, David Malkin, Anita Villani, Meredith S. Irwin, Adam Shlien.

Subscribe, Comment, Join Group

I'm interested in your feedback - please leave your comments.

To subscribe to the AI in Healthcare Milestones newsletter please click here .

To join the AI in Healthcare Milestones Group please click here .

Copyright ? 2023 Margaretta Colangelo. All Rights Reserved.

This article was written by Margaretta Colangelo. Margaretta is a leading AI analyst who tracks significant milestones in AI in healthcare. She consults with AI healthcare companies and writes about some of the companies she consults with. Margaretta serves on the advisory board of the AI Precision Health Institute at the University of Hawai?i?Cancer Center @realmargaretta