Machine Learning Effort Enables Large-Scale Cancer
Post# of 373
(Total Views: 235)
Posted On: 12/27/2022 5:09:50 PM
Machine Learning Effort Enables Large-Scale Cancer Study to Improve Boundary Detection
Scientists at the University of Pennsylvania School of Medicine and Intel Corp. carried out a large-scale international machine learning effort to collect knowledge from brain scans of more than 6,000 patients with glioblastoma at various sites globally. Their objective was to develop a model that could improve identification and prediction of boundaries in different tumor subcompartments.
Spyridon Bakas, an assistant professor at Penn Medicine, stated that the study had the single biggest and most-diverse glioblastoma patient dataset ever considered in the literature, noting that this was facilitated through federated learning. Bakas also noted that the machine learning models became more accurate when more data was fed into them, which would in turn, improve researchers’ abilities to understand, remove and treat glioblastoma in patients with increased precision.
Glioblastoma is an aggressive cancer type that occurs in the spinal cord and/or brain. Scientists studying such rare illnesses often have patient populations that are limited to their geographical locations or institutions.
Through federated learning, however, researchers can collaborate on studies without compromising patient privacy data through data sharing, which is a major obstacle. Federated learning refers to an approach that entails training a machine learning algorithm across multiple decentralized servers or devices or institutions holding local data samples, without actually exchanging the data.
Bakas, who has specialized in pathology and laboratory medicine, led this large-scale study along with first authors Sarthak Pati, Micah Sheller and Brandon Edwards of Intel Labs, along with Ujjwal Baid. The researchers’ model followed a staged approach, which began with the public initial model stage, followed by the preliminary consensus model stage and, lastly, the final consensus model stage.
The model was designed to identify boundaries of three glioblastoma tumor subcompartments: whole tumor, tumor core and enhancing tumor. Whole tumor is defined by the union of the core of the tumor and the infiltrated tissue while the tumor core includes the enhancing tumor and the part that kills the surrounding tissue. The enhancing tumor represents the vascular blood-brain barrier breakdown in the tumor.
More patient case data was added from more sites to enhance the model’s accuracy at each stage, with the researchers noting that by the final stage, the model had recorded a 27% improvement in detecting enhancing tumor boundaries; a 16% improvement in detecting whole tumor boundaries; and a 33% improvement in tumor core boundary detection.
The study was funded by the National institutes of Health. Its findings were reported in the “Nature Communications” journal.
As more insights are obtained regarding hard-to-treat cancers such as glioblastoma, entities such as CNS Pharmaceuticals Inc. (NASDAQ: CNSP) engaged in searching for better treatments for these brain and central nervous system malignancies stand a higher chance of making significant breakthroughs.
NOTE TO INVESTORS: The latest news and updates relating to CNS Pharmaceuticals Inc. (NASDAQ: CNSP) are available in the company’s newsroom at https://ibn.fm/CNSP
Please see full terms of use and disclaimers on the BioMedWire website applicable to all content provided by BMW, wherever published or re-published: http://BMW.fm/Disclaimer
Scientists at the University of Pennsylvania School of Medicine and Intel Corp. carried out a large-scale international machine learning effort to collect knowledge from brain scans of more than 6,000 patients with glioblastoma at various sites globally. Their objective was to develop a model that could improve identification and prediction of boundaries in different tumor subcompartments.
Spyridon Bakas, an assistant professor at Penn Medicine, stated that the study had the single biggest and most-diverse glioblastoma patient dataset ever considered in the literature, noting that this was facilitated through federated learning. Bakas also noted that the machine learning models became more accurate when more data was fed into them, which would in turn, improve researchers’ abilities to understand, remove and treat glioblastoma in patients with increased precision.
Glioblastoma is an aggressive cancer type that occurs in the spinal cord and/or brain. Scientists studying such rare illnesses often have patient populations that are limited to their geographical locations or institutions.
Through federated learning, however, researchers can collaborate on studies without compromising patient privacy data through data sharing, which is a major obstacle. Federated learning refers to an approach that entails training a machine learning algorithm across multiple decentralized servers or devices or institutions holding local data samples, without actually exchanging the data.
Bakas, who has specialized in pathology and laboratory medicine, led this large-scale study along with first authors Sarthak Pati, Micah Sheller and Brandon Edwards of Intel Labs, along with Ujjwal Baid. The researchers’ model followed a staged approach, which began with the public initial model stage, followed by the preliminary consensus model stage and, lastly, the final consensus model stage.
The model was designed to identify boundaries of three glioblastoma tumor subcompartments: whole tumor, tumor core and enhancing tumor. Whole tumor is defined by the union of the core of the tumor and the infiltrated tissue while the tumor core includes the enhancing tumor and the part that kills the surrounding tissue. The enhancing tumor represents the vascular blood-brain barrier breakdown in the tumor.
More patient case data was added from more sites to enhance the model’s accuracy at each stage, with the researchers noting that by the final stage, the model had recorded a 27% improvement in detecting enhancing tumor boundaries; a 16% improvement in detecting whole tumor boundaries; and a 33% improvement in tumor core boundary detection.
The study was funded by the National institutes of Health. Its findings were reported in the “Nature Communications” journal.
As more insights are obtained regarding hard-to-treat cancers such as glioblastoma, entities such as CNS Pharmaceuticals Inc. (NASDAQ: CNSP) engaged in searching for better treatments for these brain and central nervous system malignancies stand a higher chance of making significant breakthroughs.
NOTE TO INVESTORS: The latest news and updates relating to CNS Pharmaceuticals Inc. (NASDAQ: CNSP) are available in the company’s newsroom at https://ibn.fm/CNSP
Please see full terms of use and disclaimers on the BioMedWire website applicable to all content provided by BMW, wherever published or re-published: http://BMW.fm/Disclaimer
(0)
(0)