東華大學圖書館 |

Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease./
作者:	Torres-Velazquez, Maribel.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2022,
面頁冊數:	275 p.
附註:	Source: Dissertations Abstracts International, Volume: 84-01, Section: B.
Contained By:	Dissertations Abstracts International84-01B.
標題:	Biomedical engineering. -
ISBN:	9798834027034

Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease.
Torres-Velazquez, Maribel.

Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease. - Ann Arbor : ProQuest Dissertations & Theses, 2022 - 275 p.

Source: Dissertations Abstracts International, Volume: 84-01, Section: B.

Thesis (Ph.D.)--The University of Wisconsin - Madison, 2022.

This item must not be sold to any third party vendors.

Medical imaging techniques are a fundamental part of clinical workups and biomedical research as they provide a noninvasive assessment of the functionality of organs and tissues. The information acquired from medical images is often qualitative and an expert reader is needed to decipher its meaning. Hence, advanced computational methods have facilitated the extraction of useful information from the medical images to detect and assess disease. The central objective of this thesis was to explore the potential of two advanced computational methods, deep learning and radiomics, to better detect disease using magnetic resonance imaging (MRI) data. Machine Learning (ML) and radiomics are two computational techniques that allow the automatic extraction and interpretation of medical image features. Specifically, ML refers to the application of computational algorithms to perform future predictions by uncovering hidden data patterns. Furthermore, radiomics analysis is concerned with the extraction of quantitative features from image regions-of-interest (ROIs) that hold a prognosis value. Radiomic features extraction is followed by the application of statistical or ML techniques for data modeling. In this thesis, we first investigated (1) the potential of deep learning methods (a ML subfield) to detected temporal lobe epilepsy, (2) the feasibility of radiomics analysis to differentiate between subjects with intractable and not intractable epilepsy, and (3) the applicability of radiomics to detect hamstring strain injury (HSI). MRI data was used as part of each study within this thesis since it is the preferred imaging technique for the brain and muscles. Temporal lobe epilepsy (TLE) is the most common drug-resistant form of epilepsy in adults. Multiple MRI modalities are currently used for the diagnosis and characterization of TLE. Therefore, it is crucial to develop methods to more efficiently integrate multiple datasets for TLE diagnosis. In this thesis we first assessed the performance of DL and multimodal MRI data to provide an accurate classification of TLE by employing a multi-channel deep neural network (mDNN). Several mDNN models were cross-validated using brain structure metrics from structural MRI, MRI-based ROI correlation features, and personal demographic and cognitive data (PDC), individually or combined. The PDC dataset offered the most accurate classification of TLE individually and when combined with MRI-based brain structure metrics. These findings demonstrate the potential of deep learning approaches such as mDNN models to combine multiple datasets for TLE classification. Epilepsy is often treated with anti-seizure medication (ASM) but a portion of the patient population experiences a more severe form of the disease that is intractable to medication. A diagnosis of intractable epilepsy can take up to 3 years which increases the experienced clinical and social burdens. Hence, in the second study of this thesis, we investigated the feasibility of radiomics to predict medication response in epilepsy. Using T1- and T2-weighted (T1w and T2w) MRI images, radiomic features were extracted from the main four brain lobes and hippocampus and used for features selection and to cross-validate multiple support vector classifiers (SVCs). The combination of T1w and T2w-based hippocampus radiomic features offered the most accurate classification accuracy. When using T1w- and T2w-based radiomics individually, features from the hippocampus and combination from all ROIs offered the highest classification accuracy, respectively. Therefore, the potential of MRI-based radiomics to identify subjects with intractable epilepsy was demonstrated. HSIs are prevalent between collegiate and professional athletes in sprinting sports, and it is characterized by a high rate of re-injury. A premature return to sport (RTS) is associated with elevated reinjury rates. To this end, in the last study of this thesis, we explored the potential of radiomics to detect and assess HSI in collegiate athletes. Using T1w, T2w, and diffusion tensor imaging (DTI) MRI images, radiomic features were extracted from the four hamstring muscles in each limb. Feature selection and multiple SVCs were implemented to differentiate between involved and uninvolved limb and perform both binary and multiclass prediction of RTS using individual and combined multimodal MRI-based radiomics. Radiomic features from all DTI and T2w images are the strongest predictors of injury. When performing a binary and multiclass prediction, the combination of all MRI-based and all DTI maps radiomic features offered the most accurate classification, respectively. Consequently, it was demonstrated that radiomic features from hamstring muscles are related to injury and have the potential to predict RTS.

ISBN: 9798834027034Subjects--Topical Terms:

535387
Biomedical engineering.
Subjects--Index Terms:

Deep learning

Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease.
LDR:05968nmm a2200373 4500 001 2349801
005 20221003074957.5
008 241004s2022 eng d
020 $a 9798834027034
035 $a (MiAaPQ)AAI29255255
035 $a AAI29255255
040 $a MiAaPQ $c MiAaPQ
100 1 $a Torres-Velazquez, Maribel. $3 3689220
245 1 0 $a Deep Learning and Radiomics Applied to Magnetic Resonance Images of the Brain and Muscles to Better Detect Disease.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2022
300 $a 275 p.
500 $a Source: Dissertations Abstracts International, Volume: 84-01, Section: B.
500 $a Advisor: McMillan, Alan B.
502 $a Thesis (Ph.D.)--The University of Wisconsin - Madison, 2022.
506 $a This item must not be sold to any third party vendors.
520 $a Medical imaging techniques are a fundamental part of clinical workups and biomedical research as they provide a noninvasive assessment of the functionality of organs and tissues. The information acquired from medical images is often qualitative and an expert reader is needed to decipher its meaning. Hence, advanced computational methods have facilitated the extraction of useful information from the medical images to detect and assess disease. The central objective of this thesis was to explore the potential of two advanced computational methods, deep learning and radiomics, to better detect disease using magnetic resonance imaging (MRI) data. Machine Learning (ML) and radiomics are two computational techniques that allow the automatic extraction and interpretation of medical image features. Specifically, ML refers to the application of computational algorithms to perform future predictions by uncovering hidden data patterns. Furthermore, radiomics analysis is concerned with the extraction of quantitative features from image regions-of-interest (ROIs) that hold a prognosis value. Radiomic features extraction is followed by the application of statistical or ML techniques for data modeling. In this thesis, we first investigated (1) the potential of deep learning methods (a ML subfield) to detected temporal lobe epilepsy, (2) the feasibility of radiomics analysis to differentiate between subjects with intractable and not intractable epilepsy, and (3) the applicability of radiomics to detect hamstring strain injury (HSI). MRI data was used as part of each study within this thesis since it is the preferred imaging technique for the brain and muscles. Temporal lobe epilepsy (TLE) is the most common drug-resistant form of epilepsy in adults. Multiple MRI modalities are currently used for the diagnosis and characterization of TLE. Therefore, it is crucial to develop methods to more efficiently integrate multiple datasets for TLE diagnosis. In this thesis we first assessed the performance of DL and multimodal MRI data to provide an accurate classification of TLE by employing a multi-channel deep neural network (mDNN). Several mDNN models were cross-validated using brain structure metrics from structural MRI, MRI-based ROI correlation features, and personal demographic and cognitive data (PDC), individually or combined. The PDC dataset offered the most accurate classification of TLE individually and when combined with MRI-based brain structure metrics. These findings demonstrate the potential of deep learning approaches such as mDNN models to combine multiple datasets for TLE classification. Epilepsy is often treated with anti-seizure medication (ASM) but a portion of the patient population experiences a more severe form of the disease that is intractable to medication. A diagnosis of intractable epilepsy can take up to 3 years which increases the experienced clinical and social burdens. Hence, in the second study of this thesis, we investigated the feasibility of radiomics to predict medication response in epilepsy. Using T1- and T2-weighted (T1w and T2w) MRI images, radiomic features were extracted from the main four brain lobes and hippocampus and used for features selection and to cross-validate multiple support vector classifiers (SVCs). The combination of T1w and T2w-based hippocampus radiomic features offered the most accurate classification accuracy. When using T1w- and T2w-based radiomics individually, features from the hippocampus and combination from all ROIs offered the highest classification accuracy, respectively. Therefore, the potential of MRI-based radiomics to identify subjects with intractable epilepsy was demonstrated. HSIs are prevalent between collegiate and professional athletes in sprinting sports, and it is characterized by a high rate of re-injury. A premature return to sport (RTS) is associated with elevated reinjury rates. To this end, in the last study of this thesis, we explored the potential of radiomics to detect and assess HSI in collegiate athletes. Using T1w, T2w, and diffusion tensor imaging (DTI) MRI images, radiomic features were extracted from the four hamstring muscles in each limb. Feature selection and multiple SVCs were implemented to differentiate between involved and uninvolved limb and perform both binary and multiclass prediction of RTS using individual and combined multimodal MRI-based radiomics. Radiomic features from all DTI and T2w images are the strongest predictors of injury. When performing a binary and multiclass prediction, the combination of all MRI-based and all DTI maps radiomic features offered the most accurate classification, respectively. Consequently, it was demonstrated that radiomic features from hamstring muscles are related to injury and have the potential to predict RTS.
590 $a School code: 0262.
650 4 $a Biomedical engineering. $3 535387
650 4 $a Medical imaging. $3 3172799
650 4 $a Artificial intelligence. $3 516317
653 $a Deep learning
653 $a Radiomics
653 $a Magnetic resonance images
653 $a Muscles
653 $a Disease detection
653 $a Brain
690 $a 0541
690 $a 0574
690 $a 0800
710 2 0 $a The University of Wisconsin - Madison. $b Biomedical Engineering. $3 3192572
773 0 $t Dissertations Abstracts International $g 84-01B.
790 $a 0262
791 $a Ph.D.
792 $a 2022
793 $a English