Deep Learning Methods for Classification of Gliomas and Their Molecular Subtypes, From Central Learning to Federated Learning
Doctoral thesis, 2023
The most common type of brain cancer in adults are gliomas. Under the updated 2016 World Health Organization (WHO) tumor classification in central nervous system (CNS), identification of molecular subtypes of gliomas is important. For low grade gliomas (LGGs), prediction of molecular subtypes by observing magnetic resonance imaging (MRI) scans might be difficult without taking biopsy. With the development of machine learning (ML) methods such as deep learning (DL), molecular based classification methods have shown promising results from MRI scans that may assist clinicians for prognosis and deciding on a treatment strategy. However, DL requires large amount of training datasets with tumor class labels and tumor boundary annotations. Manual annotation of tumor boundary is a time consuming and expensive process.
The thesis is based on the work developed in five papers on gliomas and their molecular subtypes. We propose novel methods that provide improved performance. The proposed methods consist of a multi-stream convolutional autoencoder (CAE)-based classifier, a deep convolutional generative adversarial network (DCGAN) to enlarge the training dataset, a CycleGAN to handle domain shift, a novel federated learning (FL) scheme to allow local client-based training with dataset protection, and employing bounding boxes to MRIs when tumor boundary annotations are not available.
Experimental results showed that DCGAN generated MRIs have enlarged the original training dataset size and have improved the classification performance on test sets. CycleGAN showed good domain adaptation on multiple source datasets and improved the classification performance. The proposed FL scheme showed a slightly degraded performance as compare to that of central learning (CL) approach while protecting dataset privacy. Using tumor bounding boxes showed to be an alternative approach to tumor boundary annotation for tumor classification and segmentation, with a trade-off between a slight decrease in performance and saving time in manual marking by clinicians. The proposed methods may benefit the future research in bringing DL tools into clinical practice for assisting tumor diagnosis and help the decision making process.
Show more
The thesis is based on the work developed in five papers on gliomas and their molecular subtypes. We propose novel methods that provide improved performance. The proposed methods consist of a multi-stream convolutional autoencoder (CAE)-based classifier, a deep convolutional generative adversarial network (DCGAN) to enlarge the training dataset, a CycleGAN to handle domain shift, a novel federated learning (FL) scheme to allow local client-based training with dataset protection, and employing bounding boxes to MRIs when tumor boundary annotations are not available.
Experimental results showed that DCGAN generated MRIs have enlarged the original training dataset size and have improved the classification performance on test sets. CycleGAN showed good domain adaptation on multiple source datasets and improved the classification performance. The proposed FL scheme showed a slightly degraded performance as compare to that of central learning (CL) approach while protecting dataset privacy. Using tumor bounding boxes showed to be an alternative approach to tumor boundary annotation for tumor classification and segmentation, with a trade-off between a slight decrease in performance and saving time in manual marking by clinicians. The proposed methods may benefit the future research in bringing DL tools into clinical practice for assisting tumor diagnosis and help the decision making process.
glioma subtype classification
convolutional autoencoder
convolutional NN
multi-stream U-Net.
CycleGAN
1p/19q codeletion
federated learning
IDH mutation
generative adversarial network
Deep learning
Author
Muhaddisa Barat Ali
Chalmers, Electrical Engineering, Signal Processing and Biomedical Engineering
Included papers
Multi-stream Convolutional Autoencoder and 2D Generative Adversarial Network for Glioma Classification
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics),;Vol. 11678 LNCS(2019)p. 234-245
Paper in proceeding
Domain Mapping and Deep Learning from Multiple MRI Clinical Datasets for Prediction of Molecular Subtypes in Low Grade Gliomas
Brain Sciences,;Vol. 10(2020)p. 1-20
Journal article
Prediction of glioma‑subtypes: comparison of performance on a DL classifier using bounding box areas versus annotated tumors
BioMedical Engineering Online,;Vol. 4(2022)
Journal article
A Feasibility Study on Deep Learning Based Brain Tumor Segmentation Using 2D Ellipse Box Areas
Sensors,;Vol. 22(2022)
Journal article
A novel federated deep learning scheme for glioma and its subtype classification
Frontiers in Neuroscience,;Vol. 17(2023)
Journal article
Popular science description
English
This thesis focuses on some deep learning (DL) methods for classification of gliomas and their biomarker molecular subtypes and tumor segmentation, aimed at assisting medical doctors in their diagnostic process.
Knowing the biomarker molecular subtypes prior to surgery is of great value and provides good assistance in treatment planning. For low grade gliomas (LGGs), prediction of molecular subtypes by observing MRI scans might be difficult without taking biopsy. With the development of machine learning (ML) methods such as deep learning (DL), molecular based classification methods have shown promising results using pre-existing MRIs along with their biopsy information. Such automatic decision making processes work quicker, save the labor and provide the clinicians with tools to choose. However, DL requires large amount of training datasets with tumor class labels and tumor boundary annotations. Manual annotation of tumor boundary is a time consuming and expensive process. We propose novel methods that provide encouraging performances. The proposed methods consist of a multi-stream convolutional autoencoder (CAE)-based classifier, a deep convolutional generative adversarial network (DCGAN) to enlarge the training dataset, a CycleGAN to handle domain shift, a novel federated learning (FL) scheme to allow local client-based training with dataset protection, and employing bounding boxes to MRIs for classification and segmentation when tumor boundary annotations are not available. The proposed methods may benefit the future research in bringing DL tools into clinical practice for assisting tumor diagnosis and help the decision making process.
Knowing the biomarker molecular subtypes prior to surgery is of great value and provides good assistance in treatment planning. For low grade gliomas (LGGs), prediction of molecular subtypes by observing MRI scans might be difficult without taking biopsy. With the development of machine learning (ML) methods such as deep learning (DL), molecular based classification methods have shown promising results using pre-existing MRIs along with their biopsy information. Such automatic decision making processes work quicker, save the labor and provide the clinicians with tools to choose. However, DL requires large amount of training datasets with tumor class labels and tumor boundary annotations. Manual annotation of tumor boundary is a time consuming and expensive process. We propose novel methods that provide encouraging performances. The proposed methods consist of a multi-stream convolutional autoencoder (CAE)-based classifier, a deep convolutional generative adversarial network (DCGAN) to enlarge the training dataset, a CycleGAN to handle domain shift, a novel federated learning (FL) scheme to allow local client-based training with dataset protection, and employing bounding boxes to MRIs for classification and segmentation when tumor boundary annotations are not available. The proposed methods may benefit the future research in bringing DL tools into clinical practice for assisting tumor diagnosis and help the decision making process.
Categorizing
Subject Categories (SSIF 2011)
Computer Engineering
Human Computer Interaction
Computer Science
Computer Vision and Robotics (Autonomous Systems)
Cancer and Oncology
Other Electrical Engineering, Electronic Engineering, Information Engineering
Areas of Advance
Information and Communication Technology
Health Engineering
Identifiers
ISBN
978-91-7905-903-3
Other
Publisher
Chalmers
Public defence
2023-09-14 13:00 -- 16:00
E2 Room 3364 EDIT-rummet
Opponent: Guoying Zhao, University of Oulu, Finland.