PRISM :: Browsing by Author "Frayne, Richard"

Browsing by Author "Frayne, Richard"

Now showing 1 - 20 of 53

Open Access
A longitudinal magnetic resonance imaging study of neurodegenerative and small vessel disease, and clinical cognitive trajectories in non demented patients with transient ischemic attack: the PREVENT study
(2018-07-16) Tariq, Sana; d’Esterre, Christopher D; Sajobi, Tolulope T; Smith, Eric E; Longman, Richard S; Frayne, Richard; Coutts, Shelagh B; Forkert, Nils D; Barber, Philip A
Abstract Background Late-life cognitive decline, caused by progressive neuronal loss leading to brain atrophy years before symptoms are detected, is expected to double in Canada over the next two decades. Cognitive impairment in late life is attributed to vascular and lifestyle related risk factors in mid-life in a substantial proportion of cases (50%), thereby providing an opportunity for effective prevention of cognitive decline if incipient disease is detected earlier. Patients presenting with transient ischemic attack (TIA) commonly display some degree of cognitive impairment and are at a 4-fold increased risk of dementia. In the Predementia Neuroimaging of Transient Ischemic Attack (PREVENT) study, we will address what disease processes (i.e., Alzheimer’s vs. vascular disease) lead to neurodegeneration, brain atrophy, and cognitive decline, and whether imaging measurements of brain iron accumulation using quantitative susceptibility mapping predicts subsequent brain atrophy and cognitive decline. Methods A total of 440 subjects will be recruited for this study with 220 healthy subjects and 220 TIA patients. Early Alzheimer’s pathology will be determined by cerebrospinal fluid samples (including tau, a marker of neuronal injury, and amyloid β1–42) and by MR measurements of iron accumulation, a marker for Alzheimer’s-related neurodegeneration. Small vessel disease will be identified by changes in white matter lesion volume. Predictors of advanced rates of cerebral and hippocampal atrophy at 1 and 3 years will include in vivo Alzheimer’s disease pathology markers, and MRI measurements of brain iron accumulation and small vessel disease. Clinical and cognitive function will be assessed annually post-baseline for a period of 5-years using a clinical questionnaire and a battery of neuropsychological tests, respectively. Discussion The PREVENT study expects to demonstrate that TIA patients have increased early progressive rates of cerebral brain atrophy after TIA, before cognitive decline can be clinically detected. By developing and optimizing high-level machine learning models based on clinical data, image-based (quantitative susceptibility mapping, regional brain, and white matter lesion volumes) features, and cerebrospinal fluid biomarkers, PREVENT will provide a timely opportunity to identify individuals at greatest risk of late-life cognitive decline early in the course of disease, supporting future therapeutic strategies for the promotion of healthy aging.
Open Access
Accelerating MR Neuroimaging of Stroke Using Sparse Acquisition Coupled with Nonlinear Reconstruction Techniques
(2013-06-28) Yerly, Jerome; Frayne, Richard; Lauzon, M Louis
The guiding theme of my research is to accelerate and improve magnetic resonance (MR) imaging such that it becomes the clinical modality of choice in diagnosing, treating, and hopefully preventing stroke. Stroke, be it ischemic or haemorrhagic, is a leading cause of death and permanent disability worldwide: it is a medical emergency that requires rapid diagnosis to initiate early patient treatment and prevent irreversible brain injury. Computed tomography (CT) is currently the preferred imaging modality due to its high spatial and temporal resolution. MR imaging is a slower technique than CT, but it offers a significantly broader and more varied set of image contrasts and functional information than CT. Simply stated, the goal of my research is to accelerate the MR acquisition and/or increase resolution without sacrificing image quality in order to provide high quality diagnostic information. The most obvious way to scan faster is to acquire fewer data points, although this can often yield undesired reductions in image quality such as blurring, aliasing, or ghosting artefacts. Fortunately, numerous recent developments using multiple channel receiver coils and advanced reconstruction techniques are overcoming these drawbacks. This doctoral thesis investigates many of these advanced signal acquisition and processing techniques as they apply to stroke. In terms of diagnosis, I compare several state-of-the-art paradigms to accelerate key sequences of an acute MR stroke protocol. For treatment, I describe an enhanced passive MR catheter tracking approach that enables continuous monitoring of the catheter during endovascular procedures. And finally, with regards to stroke prevention, I present a novel imaging technique for assessing atherosclerosis in carotid arteries. In all cases, numerical and experimental verifications provided diagnostic images of very high quality (and comparable to conventional MR scans), albeit acquired 2 to 6 times faster. This work and continued efforts worldwide are inching us closer to making MR imaging the modality of choice in the comprehensive management of acute stroke patients.
Open Access
Advanced MRI methods for probing disease severity and functional decline in multiple sclerosis
(2023-12-14) Oladosu, Olayinka Adeoluwa; Zhang, Yunyan; Joshi, Manish; Dunn, Jeffrey Frank; Frayne, Richard; Le, Lawrence Trong-Huan
Multiple sclerosis (MS) is a chronic and severe disease of the central nervous system characterized by complex pathology including inflammatory demyelination and neurodegeneration. MS impacts >2.8 million people worldwide, with most starting with a relapsing-remitting form (RRMS) in young adulthood, and many of them worsening to a secondary-progressive course (SPMS) despite treatment. So, there is a clear need for improved disease characterization. MRI is an ideal tool for non-invasive assessment of MS pathology, but there is still no established measure of disease activity and functional consequences. This project aims to overcome the challenge by developing novel imaging measures based on brain diffusion MRI and phase congruency texture analysis of conventional MRI. Through advanced modeling and analysis of clinically feasible brain MRI, this thesis investigates whether and how the derived measures differentiate MS pathology types and disease severity and predict functional outcomes in MS. The overall process has led to important technical innovations in several aspects. These include: innovative modeling of simple diffusion acquisitions to generate high angular resolution diffusion imaging (HARDI) measures; new optimization and harmonization techniques for diffusion MRI; innovative neural network models to create new diffusion data for comprehensive HARDI modeling; and novel methods and a graphic user interface for optimizing phase congruency analyses. Assisted by different machine learning methods, collective findings show that advanced measures from both diffusion MRI and phase congruency are highly sensitive to subtle differences in MS pathology, which differentiate disease severity between RRMS and SPMS through multi-dimensional analyses including chronic active lesions, and predict functional outcomes especially in physical and neurocognitive domains. These results are clinically translational and the new measures and techniques can help improve the evaluation and management of both MS and similar diseases.
Open Access
An Investigation of Deep Learning Methods to Shorten GABA-edited Magnetic Resonance Spectroscopy Scan Times
(2023-09-18) Pommot Berto, Rodrigo; Medeiros de Souza, Roberto; Harris, Ashley; Lebel, Marc; Zhang, Yunyan; Frayne, Richard; McCreary, Cheryl
Edited magnetic resonance spectroscopy (MRS) can provide localized information on gamma-aminobutyric acid (GABA) concentration in vivo. However, GABA-edited MRS data has a low spectral quality, and many measurements, known as transients, need to be collected and averaged to obtain a high-quality spectrum, resulting in long scan times. This work investigated using deep learning (DL) with only a quarter of the number of conventionally acquired transients to shorten scan times by four while maintaining or improving spectral quality. A proof of concept was demonstrated by reconstructing GABA-edited spectra with only 80 transients and different configurations of DL-based pipelines. The best-performing pipeline used a proposed dimension-reducing 2D U-NET variation and it obtained better spectral quality metrics than conventionally reconstructed spectra with 320 transients. Simulated data was also shown to be useful in pre-training DL model weights. An open data challenge for reconstructing GABA-edited spectra from reduced transients was organized, and various DL models from different participating teams were compared. The challenge results reinforced the proof of concept conclusions that higher spectral quality can be achieved with DL reconstructions. However, the challenge metric evaluation also showed that DL models are able to undesirably exploit the limitations of conventional MRS metrics when using those as the training loss for the models, leading to good metric values but poor reconstructed spectra quality. DL reconstructions of GABA-edited MRS with 80 transients were also quantified and had significant differences from results from conventional reconstructions with 320 transients. However, given the lack of ground truths in the quantified data, it is not possible to conclude which results are closer to the actual concentrations. This work showed that DL methods can reduce GABA-edited MRS scan times while increasing spectral quality. Due to the lack of ground truths for the in vivo data, further studies are necessary to validate the concentrations obtained from the DL-based GABA-edited MRS reconstructions in comparison to conventional methods. This work was developed in the spirit of open science, and the data and code to reproduce the results were made publicly available.
Open Access
Assessment of T2 Magnetic Resonance Relaxation as an Imaging Biomarker of Normal Brain Aging over the Adult Lifespan
(2021-01-11) Wang, Xing; Frayne, Richard; Lebel, Catherine A.; McCreary, Cheryl R.
Quantitative T2 relaxation time (qT2) was proposed for assessing brain tissue changes. In this study, qT2 was evaluated to examine normal brain aging across the adult lifespan. I explored the specific hypotheses that 1) short- (weeks) and long-term (years) qT2 repeatability were equivalent, and 2) qT2 increased with age-related increases in tissue water content and demyelination but decreased with increased iron accumulation. The repeatability assessment found qT2 estimation robust over >4 years. Long-term was similar to, but worse than short-term repeatability. A quadratic regression model was determined to be the best fitting analytical model. Linear mixed effects models were used to evaluate qT2 changes with age in twelve regions and qT2 change rates from six regions compared against a reference region. The results supported the hypothesis that qT2 increases with age; however, could not demonstrate that these T2 changes result from changes in tissue water content, demyelination or iron accumulation.
Open Access
Background suppression using hadamard RF pulses for endovascular therapy in magnetic resonance
(2004) Nagarajappa, Nirupama; Frayne, Richard
Open Access
Better assessment of acute stroke using a revised strategy for bolus-tracking perfusion imaging and a novel volumetric analysis methodology
(2009) Kosior, Jayme Cameron; Frayne, Richard
Open Access
Better diagnoses of stroke and epilepsy with quantitative mr assessment
(2011) Kosior, Robert Karl; Frayne, Richard
Open Access
Brain Development During Childhood and Adolescence
(2016-01-15) Mah, Alyssa; Lebel, Catherine; Frayne, Richard; Wei, Xing-Chang; Forkert, Nils; Dyck, Richard
Brain development is a combination of complex physiological changes, and various magnetic resonance imaging (MRI) techniques can help explain observed changes during development in vivo. Building upon observations from post-mortem studies, advancements in imaging and modelling techniques provide new means to further interpret the understanding of healthy brain development during childhood and adolescence. It is, however, a challenge to capture specific physiological changes, such as myelination, using MRI. This thesis uses MRI techniques – neurite orientation dispersion and density imaging (NODDI), inhomogenous magnetization transfer (ihMT), and multi-component driven equilibrium single pulse observation of T1 and T2 (mcDESPOT) – that further characterize development in white and subcortical grey matter regions in the brain by improving specificity of the MRI signal compared to conventional techniques. Measures from NODDI, ihMT, and mcDESPOT suggest an increase in myelination and/or axonal packing during development from 0-13 years.
Open Access
Characterization of OA Severity in Knee Articular Cartilage in-Vivo Using MR Imaging and Loading Techniques
(2017) Dai, Xu; Ronsky, Janet Lenore; Frayne, Richard; Schmidt, Tannin; Boyd, Steven Kyle; Nowicki, Edwin Peter; Holdsworth, David W.
Early osteoarthritis (OA) is primarily associated with proteoglycan (PG) loss and changes in collagen structure. MR T2 imaging of knee under in-vivo loading may help to further reveal the differences between healthy and OA cartilage. This study investigated the in-vivo loading effect on MR T2 values of human knee patellar cartilage. The results demonstrated T2 value distributions in patellar cartilage were inhomogeneous. In-vivo loading had a site-specific influence on participants’ T2 values. The in-vivo loading produced a significant difference on T2 values in the middle region of interest (ROI) of patellar cartilage (p=0.004＜0.025), but not at the superior or inferior ROIs. The T2 value variation for the OA group during loading was lower than that of the healthy group (p=0.016＜0.025). The T2 recovery ratio was presented in this study as a new variable. The findings indicated the T2 recovery ratio of the OA group was significantly lower than the healthy group (p=0.042＜0.05) in the patellar cartilage middle ROI. It suggests that the OA cartilage had weaker ability to restore its original status after off-loading than healthy cartilage. This study examined the glycosaminoglycan (GAG) mass% concentration (relating PG) in human cadaveric patellar cartilages using biochemical assay. Results showed that the GAG mass% concentrations in OA lesion positions were lower than that in comparative healthy positions (p＜0.001). MR T2 imaging of healthy and OA cadaver knee joints were performed. Correlations between T2 values and the GAG mass(%) of cadaver patellar cartilage specimens were established. As PG concentration of in-vivo human articular cartilage cannot be directly measured non-invasively, the correlation of cadaveric patellar cartilage may serve as an important bridge between the T2 value and GAG mass% for living human assessment. The findings provide an indirect approach to estimate PG concentration of in-vivo patellar cartilage based on an individual’s cartilage T2 values to evaluate the extent of degradation within cartilage. This subject specific method is especially suitable for longitudinal evaluation of OA. By position-matched comparison of previous and current T2 images, the GAG mass% variation may be estimated to assess OA progression non-invasively.
Open Access
Characterizing acute ischemic stroke evolution using magnetic resonance diffusion imaging
(2008) Harris, Ashley D.; Frayne, Richard
Open Access
Combined Segmentation and Hemodynamic Analysis of Cerebrovascular Structures using Arterial Spin Labeling
(2020-01-17) Phellan Aro, Renzo; Forkert, Nils Daniel; Frayne, Richard; Walker, Richard E. A.; Lebel, Robert Marc; Far, Behrouz H.; Duong, Luc
Spatiotemporal arterial spin labeling magnetic resonance angiography (4D ASL MRA) is a non-invasive imaging modality used to acquire dynamic images of cerebrovascular structures. It can achieve high spatial and temporal resolution, while capturing morphological and blood flow data. Recent scientific studies have used 4D ASL MRA to analyze the cerebrovascular system for characterization, diagnosis, and post-treatment assessment of different cerebrovascular diseases, such as aneurysms, arteriovenous malformations, and moyamoya disease. However, this image sequence generates a considerable amount of data, which can be tedious to analyze by simple visual inspection, a problem also present with other 4D imaging methods. In this case, medical image processing methods can be used to extract the morphological and blood flow data contained in 4D ASL MRA datasets and present it in a more useful format to clinicians and researchers. The aim of this work was to develop and evaluate novel image processing methods for advanced analysis of 4D MRA datasets. The overreaching idea for the development of the corresponding methods is to use blood flow information for improving the vessel segmentation while the vessel segmentation is used to improve the results of the hemodynamic analysis. It was hypothesized that this combined analysis improves the vessel segmentation and results the hemodynamic analysis at the same time. The methods were developed and evaluated using 15 datasets of healthy volunteers, flow phantom measurements, and two datasets of patients with a stenosis. The findings of this work indicate that the proposed combined segmentation and hemodynamic analysis can improve the overall accuracy of the segmentation and blood flow parameter estimation while first experiments also show that the proposed methods can be applied to patients with a cerebrovascular disease. The methods developed in this work could help translating 4D ASL MRA datasets into clinical practice and support clinical research of various cerebrovascular diseases using 4D ASL MRA while the developed methods have also the potential to be useful for other 4D imaging sequences.
Open Access
Computational fluid dynamics as aid tool for the management of aortic wall diseases
(2019-10-24) Forneris, Arianna; Di Martino, Elena S.; Frayne, Richard; Wood, David H.
Aortic aneurysms and dissections are pathological conditions affecting the aorta. Despite being different in clinical presentation, these pathologies share a high mortality rate, as well as a lack of reliable prognostic predictors. Local fluid dynamics is assumed to play a role in aortic patho-physiology and to be a key factor responsible for aortic weakening and expansion. In this context, the numerical modeling of aortic hemodynamics, by means of image-based computational fluid dynamics (CFD), gives access to patient-specific flow-related information that may complement medical imaging in the assessment of individual aortas and support outcomes prediction. Moreover, the deformability of the aortic wall appears to be related to its strength: areas at elevated strain may, therefore, indicate structural weakening. Based on these understandings, this research work proposes the use of hemodynamic descriptors, derived from CFD simulations, to correlate local altered flow patterns with aortic remodeling and weakening, and ultimately help defining a rationale for improved rupture risk stratification. Wall shear stress-based hemodynamic descriptors were used to retrospectively assess a population of uncomplicated type B aortic dissections (ADs) with known individual outcomes. The effect of rigid versus moving wall assumption on aortic flow patterns was explored by means of fluid-structure interaction (FSI) simulation. The results highlighted the need for a patient-tailored approach when evaluating ADs, and showed the potential of CFD-derived hemodynamics to complement anatomical assessment and assist outcomes prediction. The inclusion of wall motion in the simulation of a type B AD, led to differences in value for the hemodynamic wall descriptors, however, regions of interest with respect to altered flow patterns were consistently localized by both the CFD and FSI models. Finally, a combined CFD and in-vivo strain analysis approach was developed to assess local weakening and rupture risk for a population of AAAs. A novel index, Regional Rupture Potential, was defined and proved able to capture aortic regional weakening. This thesis demonstrated the importance of accessing hemodynamic information when assessing individual aortas with prognostic purposes, along with the potential of a novel combined approach to improve aortic assessment for risk stratification.
Open Access
Computer-assisted Screening of Motion Artefact for Quality Control in Large-scale MR Imaging Trials
(2017) Adair, David; Frayne, Richard; Gobbi, David; Hu, Yaoping; Krishnamurthy, Diwakar
As the scale of medical imaging trials increases, manual quality control of the enormous volume of imaging data becomes intractable and costly. Machine learning may provide solutions to reduce the challenge of these large trials through the development of computer-assisted screening tools. The objective of this dissertation was to evaluate the suitability of machine learning for solving scalability problems of manual quality control by training an automated classifier to detect simulated motion artefact on otherwise high-quality magnetic resonance images of healthy human brain. The classifier achieved high accuracy (98.5%) without any performance optimization, and, incidentally, discovered a screening error within the experiment dataset, further demonstrating the power of machine learning in this domain and encouraging further research towards computer-assisted screening tools.
Open Access
Decreasing Brain Functional Network Segregation with Healthy Aging
(2023-03-20) Singh Sidhu, Abhijot; Frayne, Richard; Goodyear, Bradley; Bray, Signe; McCreary, Cheryl
The functional architecture of the human brain consists of distinct sensory and associative functional networks that interact as needed. Resting-state functional magnetic resonance imaging (rs-fMRI) has shown that functional connections (connectivity) between networks strengthen with age, suggesting that networks reconfigure in healthy aging by becoming less segregated. Few studies, however, have replicated these findings or investigated sex differences in network organization, and no studies have thoroughly investigated if age-related differences involve certain networks more than others. It is crucial to better understand healthy age-related brain changes as it provides a foundation to better address and investigate age-related diseases like dementia or Alzheimer’s disease. In this thesis, cross-sectional functional and structural MR data from 231 presumed healthy adult participants (~59% self-reported as female, aged 18-91 y) from the Calgary Normative Study were examined to investigate the hypotheses that (1) whole-brain and network segregation decreases with age due to decreasing within-network and increasing between-network connectivity, and (2) age-associated increases in between-network functional connectivity occur in associative network connections. Structural and functional MR data were parcellated into seven known functional networks, and average within- and between-network connectivity were computed as the z-transformed Pearson cross-correlation coefficients (zw and zb, respectively). Whole-brain and network segregation were then computed using the segregation index, SI = (|zw|-|zb|)/(|zw|). All networks exhibited decreasing segregation with age, because of increasing between-network connectivity; however, within-network connectivity did not change with age. Further, between-network connectivity increased with age in sensory-to-associative and associative-to-associative network connections, as hypothesized. Increased connectivity between associative networks was also observed in males, regardless of age. This thesis demonstrates that functional networks reconfigure with age because of increased connectivity in sensory-to-associative and associative-to-associative network connections. This may constitute compensatory and/or de-differentiation processes as humans age. The findings of this thesis also provide a foundation for better interpretation of changes in brain function that occur in age-related diseases.
Open Access
Deep Learning for Domain-Invariant Magnetic Resonance Carotid Artery Wall Segmentation
(2019-04-26) Danko, Anna M.; Frayne, Richard; Souza, Roberto Medeiros De; Rittner, Leticia; Zhang, Yunyan
Segmentation of the carotid arteries is a prerequisite to image processing techniques that are applied to medical images to assess the features of atherosclerosis, a disease which can lead to ischemic stroke. Carotid artery segmentation is currently mainly done manually in a time-consuming processing. In this work deep learning approaches were applied to carotid artery segmentation. Additionally, the influence of image contrast on segmentation performance was explored, and whether a network could be taught to learn domain-invariant features including the use of adversarial methods. Non-adversarial and adversarial methods were successfully demonstrated.
Open Access
Deep-learning-based Multi-visit Magnetic Resonance Imaging Reconstruction: Proof of Concept and Robustness Evaluation on a Cohort of Glioblastoma Patients
(2023-01-19) Beauferris, Youssef; Medeiros de Souza, Roberto; Frayne, Richard; Fear, Elise
Magnetic Resonance (MR) imaging is a powerful imaging technique for assessing brain-related diseases. However, MR scans suffer from long acquisition times and as a consequence, patients in Canada must wait extensive periods for access to a scanning session. Compressed Sensing (CS) and Parallel Imaging (PI) are two proven techniques employed to enable accelerated acquisitions. However, they both require complex reconstruction algorithms which disable real-time results. The renewed advent of deep-learning has helped tackle this problem of long reconstruction times. But, currently deep-learning based reconstruction methods do not leverage the wealth of mutual information contained across multiple patient visits to the scanner. This led to the proposal of the Multi-visit Integration Model (MIM) which is a framework for reconstructing a follow-up scan, that has been aggressively undersampled, by leveraging a previous scan. This thesis aims to investigate the performance of the MIM when similarity is not guaranteed between the previous and follow-up scan, such as in the case of glioblastoma patients. The results demonstrated that the MIM leaves localized regions, which have undergone a structural change from one scan to the next, the same. However, this conservative behaviour is not demonstrated during our robustness analysis when synthetic lesions are added to the previous scan to simulate a structural change. The effect of the single-visit reconstruction model on the multi-visit reconstruction performance demonstrated that regardless of the model used, statistically significant improvements to reconstruction quality were observed after multi-visit integration. Multi-visit reconstruction produced using older scans compared to newer scans was found to be of lower quality but still did not introduce biases towards the previous time-point. Finally, the accumulation of system error when using a multi-visit reconstruction as a previous scan in the MIM was minimal. This investigation provided insight into the behaviour of multi-visit integration in the face of structural brain changes and paves the early road towards clinical adoption.
Open Access
DeepCADe: A Deep Learning Architecture for the Detection of Lung Nodules in CT Scans
(2018-01-16) Golan, Rotem; Jacob, Christian; Denzinger, Joerg; Gavrilova, Marina; Frayne, Richard; Cunningham, Ian
Early detection of lung nodules in thoracic Computed Tomography (CT) scans is of great importance for the successful diagnosis and treatment of lung cancer. Due to improvements in screening technologies, and an increased demand for their use, radiologists are required to analyze an ever increasing amount of image data, which can affect the quality of their diagnoses. Computer-Aided Detection (CADe) systems are designed to assist radiologists in this endeavor. In this thesis, we present DeepCADe, a novel CADe system for the detection of lung nodules in thoracic CT scans which produces improved results compared to the state-of-the-art in this field of research. CT scans are grayscale images, so the terms scans and images are used interchangeably in this work. DeepCADe was trained with the publicly available Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) database, which contains 1018 thoracic CT scans with nodules of different shape and size, and is built on a Deep Convolutional Neural Network (DCNN), which is trained using the backpropagation algorithm to extract volumetric features from the input data and detect lung nodules in sub-volumes of CT images. Considering only lung nodules that have been annotated by at least three radiologists, DeepCADe achieves a 2.1% improvement in sensitivity (true positive rate) over the best result in the current published scientific literature, assuming an equal number of false positives (FPs) per scan. More specifically, it achieves a sensitivity of 89.6% with 4 FPs per scan, or a sensitivity of 92.8% with 10 FPs per scan. Furthermore, DeepCADe is validated on a larger number of lung nodules compared to other studies (Table 5.2). This increases the variation in the appearance of nodules and therefore makes their detection by a CADe system more challenging. We study the application of Deep Convolutional Neural Networks (DCNNs) for the detection of lung nodules in thoracic CT scans. We explore some of the meta parameters that affect the performance of such models, which include: 1. the network architecture, i.e. its structure in terms of convolution layers, fully-connected layers, pooling layers, and activation functions, 2. the receptive field of the network, which defines the dimensions of its input, i.e. how much of the CT scan is processed by the network in a single forward pass, 3. a threshold value, which affects the sliding window algorithm with which the network is used to detect nodules in complete CT scans, and 4. the agreement level, which is used to interpret the independent nodule annotations of four experienced radiologists. Finally, we visualize the shape and location of annotated lung nodules and compare them to the output of DeepCADe. This demonstrates the compactness and flexibility in shape of the nodule predictions made by our proposed CADe system. In addition to the 5-fold cross validation results presented in this thesis, these visual results support the applicability of our proposed CADe system in real-world medical practice.
Open Access
Developing TomoNet: A Deep Learning Feature Extractor for Medical Image Classification
(2023-06) Guerra - Librero Camacho, Javier; Frayne, Richard; Pinheiro Bento, Mariana; Pichardo, Samuel; Zhang, Yunyan; Garcia Flores, Julio Garcia; Whelan, Patrick
In this thesis, I investigated whether incorporating medical images in the training of feature extractors would enhance their performance in medical imaging tasks. To achieve this objective, I conducted two studies. In the first study, I compared the performance of a feature extractor pretrained with natural images from ImageNet (referred to as ImageNet-derived) with a feature extractor trained from scratch without any pre-existing feature extraction capabilities. Both feature extractors were trained to classify medical images into nineteen categories. The results demonstrated that the performance of the feature extractor trained from scratch approached the performance of the ImageNet-derived feature extractor as the size of the training set increased. Additionally, I introduced a metric called the "transfer learning gap" to quantify the relative amount of knowledge transfer in transfer learning. The experiments revealed that training the feature extractor twice, first with the ImageNet dataset and then with 160,000 medical images, resulted in a better-performing deep learning model. This was an unexpected result and likely due to training with an insufficient number of medical images. Thus, I named the better performing, twice-trained feature extractor "TomoNet" and used it in my subsequent studies. In the second study, I compared the performance of TomoNet with that of the ImageNet-derived feature extractor in classifying medical images based on sex. This study failed to provide conclusive evidence that TomoNet, the twice-trained feature extractor, outperformed the ImageNet-derived feature extractor. Both models exhibited signs of overfitting, and there was potential evidence of catastrophic interference in TomoNet results. Finally, I presented an initial business model for a company aiming to commercialize products derived from the TomoNet studies. The business model focused on the commercialization of "Petal-Blue," a deep learning model designed to detect white matter hyperintensities associated with dementia in brain imaging. The Canvas Business Model framework was utilized to evaluate crucial aspects of the startup, including value proposition, customer segments, revenue streams, and partnerships.
Open Access
Development, characterization and application of the s-transform to medical imaging
(2008) Brown, Robert A.; Frayne, Richard
The S-transform (ST) is a relatively new technique, first proposed in 1996 and applied to problems in geophysics. The properties and applications of the transform, and its relationship to existing techniques have not been fully explored. Though some applications, particularly in medical signal and image processing, have shown promise, the high computational requirements of the algorithm have hampered research and application of the ST. In this thesis the computational requirements of the ST are addressed and a promising image processing technique using the transform is characterized, validated and applied to several biomedical imaging problems. Two approaches to speeding up the ST are explored: (1) a parallel ST algorithm is introduced, which allows the ST calculation to take advantage of modern multi-processor hardware and computing clusters and (2) a fast ST algorithm is formulated, which dramatically reduces the computational requirements of the transform. Additionally, a general description of time-frequency transforms is presented, which clearly demonstrates the similarities and differences between the ST and both the wavelet and Fourier families of transforms. Extending this theoretical relationship, a technique that uses the ST to detect and quantify image texture is compared to an equivalent procedure based on the Fourier transform, and validated using a purpose built magnetic resonance phantom. The ST texture analysis technique is then applied to several examples of medical image analysis problems. One of these potential applications, detection of an important genetic marker in brain tumours, is explored in depth with a clinical study of 54 patients.