In this Ph.D. thesis, we develop effective techniques for medical image classification, particularly, for histopathological and magnetic resonance images (MRI). Our techniques are capable of handling the high variability in the content of such images. Handcrafted techniques based on texture analysis are used for the classification task. We also use deep learning models but training such models from scratch can be a challenging process, instead, we employ deep features and transfer learning. First, we propose a combined texture-based feature representation that is computed in the complex shearlet domain for histopathological image classification. With complex coefficients, we examine both the magnitude and relative phase of shearlets to form the feature space. Our proposed techniques are successful for histopathological image classification. Furthermore, we investigate their ability to generalize to MRI datasets that present an additional challenge, namely high dimensionality. An MRI sample consists of a large number of slices. Our proposed shearlet-based feature representation for histopathological images cannot be used without adjustment. Therefore, we consider the 3D shearlet transform given the volumetric nature of MRI data. An advantage of the 3D shearlet transform is that it takes into consideration adjacent slices of MRI data. Secondly, we study the classification of histopathological images using pre-trained deep learning models. A pre-trained deep learning model can act as a starting point for datasets with a limited number of samples. Therefore, we used various models either as unsupervised feature extractors, or weight initializers to classify histopathological images. When it comes to MRI samples, fine-tuning a deep learning model is not straightforward. Pre-trained models are trained on RGB images which have a channel size of 3, but an MRI sample has a larger number of slices. Fine-tuning a convolutional neural network (CNN) requires adjusting a model to work with MRI data. We fine-tune pre-trained models and then use them as feature extractors. Thereafter, we demonstrate the effectiveness of fine-tuned deep features with classical machine learning (ML) classifiers, namely a support vector machine and a decision tree bagger. Furthermore, instead of using a classical ML classifier for the MRI sample, we built a custom CNN that takes both the 3D shearlet descriptors and deep features as an input. This custom network processes our feature representation end-to-end and then classifies an MRI sample. Our custom CNN is more effective in comparison to a classical ML on a hidden MRI dataset. It is an indication that our CNN model is less susceptible to over-fitting.

Polymicrogyria (PMG) is one brain disease that mainly occurs in the pediatric brain. Heavy PMG will cause seizures, delayed development, and a series of problems. For this reason, it is critical to effectively identify PMG and start early treatment. Radiologists typically identify PMG through magnetic resonance imaging scans. In this study, we create and open a pediatric MRI dataset (named PPMR dataset) including PMG and controls from the Children's Hospital of Eastern Ontario (CHEO), Ottawa, Canada. The difference between PMG MRIs and control MRIs is subtle and the true distribution of the features of the disease is unknown. Hence, we propose a novel center-based deep contrastive metric learning loss function (named cDCM Loss) to deal with this difficult problem. Cross-entropy-based loss functions do not lead to models with good generalization on small and imbalanced dataset with partially known distributions. We conduct exhaustive experiments on a modified CIFAR-10 dataset to demonstrate the efficacy of our proposed loss function compared to cross-entropy-based loss functions and the state-of-the-art Deep SAD loss function. Additionally, based on our proposed loss function, we customize a deep learning model structure that integrates dilated convolution, squeeze-and-excitation blocks and feature fusion for our PPMR dataset, to achieve 92.01% recall. Since our suggested method is a computer-aided tool to assist radiologists in selecting potential PMG MRIs, 55.04% precision is acceptable. To our best knowledge, this research is the first to apply machine learning techniques to identify PMG only from MRI and our innovative method achieves better results than baseline methods.

Computed Tomography (CT) is often used for building 3D biomechanical models of human anatomy. This method exposes the subject to a significant x-ray dose and provides limited soft-tissue contrast. Magnetic Resonance Imaging (MRI) is a potential alternative to CT for this application, as MRI offers significantly better soft-tissue contrast and does not expose the subject to ionizing radiation. However, MRI requires long scan times to achieve 3D images at sufficient resolution, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). These long scan times can make subject motion a problem. This thesis describes my work to reduce scan time while achieving sufficient resolution, SNR, and CNR for 3D biomechanical modeling of (1) the human larynx, and (2) the human hip. I focused on two important strategies for reducing scan time and improving SNR and CNR: the design of RF coils optimized to detect MRI signals from the anatomy of interest, and the determination of MRI relaxation properties of the tissues being imaged (allowing optimization of imaging parameters to improve CNR between tissues). Work on the larynx was done in collaboration with the Thomson group in Mechanical Engineering at BYU. To produce a high-resolution 3D image of the larynx, a 2-channel phased array was constructed. Eight different coil element designs were analyzed for use in the array, and one chosen that provided the highest Q-ratio while still meeting the mechanical constraints of the problem. The phased array was tested by imaging a pig larynx, a good substitute for the human larynx. Excellent image quality was achieved and MR relaxometry was then performed on tissues in the larynx. The work on the hip was done in collaboration with the Anderson group in orthopedics at the University of Utah, who are building models of femoral acetabular impingement (FAI). Accurate imaging of hip cartilage requires injection of fluid into the hip joint capsule while in traction. To optimize contrast, MR relaxometry measurements were performed on saline, isovue, and lidocaine solutions (all typically injected into the hip). Our analysis showed that these substances actually should not be used for MR imaging of the hip, and alternate strategies should be explored as a result.

[ES] Según la Organización Mundial de la Salud (OMS), aproximadamente mil millones de perso- nas en todo el mundo están afectadas por trastornos neurológicos periféricos y centrales, incluidos tumores cerebrales, enfermedad de Parkinson (EP), enfermedad de Alzheimer (EA), esclerosis múl- tiple (EM), epilepsia, demencia, enfermedades neuroinfecciosas, accidentes cerebrovasculares y lesiones cerebrales traumáticas. El accidente cerebrovascular isquémico y la "enfermedad de Alz- heimer con otras demencias" son la segunda y la quinta causa principal de muerte, respectivamente. En este contexto, la detección y clasificación de lesiones cerebrales constituye un área crítica de investigación en el procesamiento de imágenes médicas, impactando significativamente tanto en la práctica clínica como en el avance científico Este proyecto tiene como objetivo proponer, desarrollar e implementar un método para la de- tección y clasificación de lesiones isquémicas y enfermedades desmielinizantes en imágenes de resonancia magnética (MRI), las cuales constituyen un tipo de hiperintensidades de la sustancia blanca (WMH). Esta tarea es crucial debido a la similitud entre estas dos enfermedades; un diagnóstico erróneo por parte de un médico no capacitado o inexperto podría llevar a un tratamiento incorrecto. Por lo tanto, este proyecto busca proporcionar a la comunidad científica y clínica una herramienta que ayude en el diagnóstico de estas enfermedades, sirviendo como una segunda opinión y como un recurso de capacitación para la identificación de lesiones cerebrales El proyecto emplea diversas técnicas de aprendizaje automático y aprendizaje profundo para comprender las características de las lesiones (biomarcadores) y facilitar su detección y clasificación. Dada la limitada cantidad de datos disponibles para el desarrollo de algoritmos, se utilizaron varios enfoques de aprendizaje por transferencia, aumento de datos clásico y aumento de datos sintético. La metodología para la detección y clasificación involucró principalmente los siguientes modelos: U-Net, Segmenting Anything Model, YOLOv8 y Detectron2. Además, se propuso un modelo utilizando redes ResNet18 para la clasificación de Regiones de Interés (ROIs). Los resultados experimentales indicaron que el modelo U-Net logró un coeficiente Dice medio de 0.95 para la segmentación. El modelo Detectron2 demostró una precisión de 0.98 en la detección y de 0.93 en la clasificación de lesiones, incluidas las lesiones pequeñas donde otros modelos a menudo fallan. El clasificador de ROIs logró una precisión de clasificación de 0.96. Estos resulta- dos sugieren que los modelos propuestos podrían ser evaluados más a fondo en un entorno clínico para mejorar su rendimiento con más datos. En general, los métodos desarrollados en este proyecto presentan un marco robusto para la detección y clasificación de lesiones cerebrales utilizando téc- nicas avanzadas de aprendizaje automático. Los hallazgos indican que los modelos desarrollados podrían ayudar significativamente en los diagnósticos clínicos, proporcionando un apoyo confiable para los médicos y contribuyendo a mejores resultados para los pacientes. / [CA] Segons l'Organització Mundial de la Salut (OMS), aproximadament mil milions de persones a tot el món estan afectades per trastorns neurològics perifèrics i centrals, incloent tumors cerebrals, malaltia de Parkinson (EP), malaltia d'Alzheimer (EA), esclerosi múltiple (EM), epilèpsia, demència, malalties neuroinfeccioses, accidents cerebrovasculars i lesions cerebrals traumàtiques. L'accident cerebrovascular isquèmic i la "malaltia d'Alzheimer amb altres demències" són la segona i la cinquena causa principal de mort, respectivament. En aquest context, la detecció i classificació de lesions cerebrals constitueix una àrea crítica d'investigació en el processament d'imatges mèdiques, impactant significativament tant en la pràctica clínica com en l'avenç científic. Aquest projecte té com a objectiu proposar, desenvolupar i implementar un mètode per a la detecció i classificació de lesions isquèmiques i malalties desmielinitzants en imatges de ressonància magnètica (MRI), les quals constitueixen un tipus d'hiperintensitats de la substància blanca (WMH). Aquesta tasca és crucial a causa de la similitud entre aquestes dues malalties; un diagnòstic erroni per part d'un metge no capacitat o inexpert podria portar a un tractament incorrecte. Per tant, aquest pro- jecte busca proporcionar a la comunitat científica i clínica una eina que ajude en el diagnòstic d'aquestes malalties, servint com una segona opinió i com un recurs de capacitació per a la identificació de lesions cerebrals. El projecte empra diverses tècniques d'aprenentatge automàtic i aprenentatge profund per a com- prendre les característiques de les lesions (biomarcadors) i facilitar la seua detecció i classificació. Donada la limitada quantitat de dades disponibles per al desenvolupament d'algoritmes, s'utilitzaren diversos enfocaments d'aprenentatge per transferència, augment de dades clàssic i augment de dades sintètic. La metodologia per a la detecció i classificació va involucrar principalment els següents mo- dels: U-Net, Segmenting Anything Model, YOLOv8 i Detectron2. A més, es va proposar un model utilitzant xarxes ResNet18 per a la classificació de Regions d'Interès (ROIs). Els resultats experimentals indicaren que el model U-Net va aconseguir un coeficient Dice mitjà de 0.95 per a la segmentació. El model Detectron2 va demostrar una precisió de 0.98 en la detecció i de 0.93 en la classificació de lesions, incloent les lesions petites on altres models sovint fallen. El classificador de ROIs va aconseguir una precisió de classificació de 0.96. Aquests resultats suggereixen que els models proposats podrien ser avaluats més a fons en un entorn clínic per a millorar el seu rendiment amb més dades. En general, els mètodes desenvolupats en aquest projecte presenten un marc robust per a la detecció i classificació de lesions cerebrals utilitzant tècniques avançades d'aprenentatge automàtic. Els descobriments indiquen que els models desenvolupats podrien ajudar significativament en els diagnòstics clínics, proporcionant un suport fiable per als metges i contribuint a millors resultats per als pacients. / [EN] According to the World Health Organization (WHO), approximately one billion people worldwide are affected by peripheral and central neurological disorders, including brain tumors, Parkinson's disease (PD), Alzheimer's disease (AD), multiple sclerosis (MS), epilepsy, dementia, neuroinfectious diseases, stroke, and traumatic brain injuries. Ischemic stroke, AD, and other dementias are the second and fifth leading causes of death, respectively. In this context, detecting and classifying brain lesions constitute a critical area of research in medical imaging processing, significantly impacting both clinical practice and scientific advancement. This project aims to develop and implement a method for detecting and classifying ischemic lesions and demyelination diseases in MRI images, which are identified by White Matter Hyperintensities. This task is crucial due to the similarity between these two diseases; a misdiagnosis by an untrained or inexperienced physician could lead to incorrect treatment. Therefore, this project seeks to provide the scientific and clinical community with a tool that assists in diagnosing these diseases, serving as a second opinion and as a training resource for identifying brain lesions. The project employs machine learning and deep learning techniques to understand lesion features (biomarkers) and facilitate detection and classification. Given the amount of data available for algorithm development, several transfer learning approaches, classical data augmentation, and synthetic data augmentation methods were utilized. The methodology for detection and classification primarily involved the following models: U-Net, Segmenting Anything Model, YOLOv8, and Detectron2. Additionally, a model using ResNet18 networks was proposed to classify Regions of Interest (ROIs). Experimental results indicated that the U-Net model achieved a mean Dice coefficient of 0.94 for segmentation. The Detectron2 model demonstrated an accuracy of 0.98 in detecting and 0.93 in classifying lesions, including small lesions where other models often fail. The ROI classifier using Res- Net18 achieved a classification accuracy of 0.96. These results suggest that the proposed models could be further evaluated in a clinical environment to enhance their performance. In conclusion, the methods developed in this thesis and the findings indicate that they could significantly aid in clinical diagnostics, providing reliable support for physicians and contributing to better patient outcomes. / Castillo Malla, DP. (2024). Magnetic Resonance Image Algorithm to Identify Demyelination and Ischemia Brain Diseases [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/211187