Advances in MR technology have improved the potential for visualization of small lesions in brain images. This has resulted in the opportunity to detect cerebral microbleeds (CMBs), small hemorrhages in the brain that are known to be associated with risk of ischemic stroke and intracerebral bleeding. Currently, no computerized method is available for fully- or semi-automated detection of CMBs. In this paper, we propose a CAD system for the detection of CMBs to speed up visual analysis in population-based studies. Our method consists of three steps: (i) skull-stripping (ii) initial candidate selection (iii) reduction of false-positives using a two layer classi cation and (iv) determining the anatomical location of CMBs. The training and test sets consist of 156 subjects (448 CMBs) and 81 subjects (183 CMBs), respectively. The geometrical, intensity-based and local image descriptor features were used in the classi cation steps. The training and test sets consist of 156 subjects (448 CMBs) and 81 subjects (183 CMBs), respectively. The sensitivity for CMB detection was 90% with, on average, 4 false-positives per subject.

Quantitative cancer imaging is an emerging field that develops computational techniques to acquire a deep understanding of cancer characteristics for cancer diagnosis and clinical decision making. The recent emergence of growing clinical imaging data provides a wealth of opportunity to systematically explore quantitative information to advance cancer diagnosis. Crucial questions arise as to how we can develop specific computational models that are capable of mining meaningful knowledge from a vast quantity of imaging data and how to transform such findings into improved personalized health care? This dissertation presents a set of computational models in the context of malignant brain tumors— Giloblastoma Multiforme (GBM), which is notoriously aggressive with a poor survival rate. In particular, this dissertation developed quantitative feature extraction approaches for tumor diagnosis from magnetic resonance imaging (MRI), including a multi-scale local computational feature and a novel regional habitat quantification analysis of tumors. In addition, we proposed a histogram-based representation to investigate biological features to characterize ecological dynamics, which is of great clinical interest in evaluating tumor cellular distributions. Furthermore, in regards to clinical systems, generic machine learning techniques are typically incapable of generalizing well to specific diagnostic problems. Therefore, quantitative analysis from a data-driven perspective is becoming critical. In this dissertation, we propose two specific data-driven models to tackle different types of clinical MRI data. First, we inspected cancer systems from a time-domain perspective. We propose a quantitative histogram-based approach that builds a prediction model, measuring the differences from pre- and post-treatment diagnostic MRI data. Second, we investigated the problem of mining knowledge from a skewed distribution—data samples of each survival group are unequally distributed. We proposed an algorithmic framework to effectively predict survival groups by jointly considering imbalanced distributions and classifier design. Our approach achieved an accuracy of 95.24%, suggesting it captures class-specific information in a challenging clinical setting.

Microscopy image processing is an emerging and quickly growing field in medical imaging research area. Recent advancements in technology including higher computation power, larger and cheaper storage modules, and more efficient and faster data acquisition devices such as whole-slide imaging scanners contributed to the recent microscopy image processing research advancement. Most of the methods in this research area either focus on automatically process images and make it easier for pathologists to direct their focus on the important regions in the image, or they aim to automate the whole job of experts including processing and classifying images or tissues that leads to disease diagnosis. This dissertation is consisted of four different frameworks to process microscopy images. All of them include methods for segmentation either as the whole suggested framework or the initial part of the framework for future feature extraction and classification. Specifically, the first proposed framework is a general segmentation method that works on histology images from different tissues and segments relatively solid nuclei in the image, and the next three frameworks work on cervical microscopy images, segmenting cervical nuclei/cells. Two of these frameworks focus on cervical tissue segmentation and classification using histology images and the last framework is a comprehensive segmentation framework that segments overlapping cervical cells in cervical cytology Pap smear images. One of the several commonalities among these frameworks is that they all work at the region level and use different region features to segment regions and later either expand, split or refine the segmented regions to produce the final segmentation output. Moreover, all proposed frameworks work relatively much faster than other methods on the same datasets. Finally, proving ground truth for datasets to be used in the training phase of microscopy image processing algorithms is relatively time-consuming, complicated and costly. Therefore, I designed the frameworks in such a way that they set most (if not all) of the parameters adaptively based on each image that is being processed at the time. All of the included frameworks either do not depend on training datasets at all (first three of the four discussed frameworks) or need very small training datasets to learn or set a few parameters.

abstract: Traumatic injuries are the leading cause of death in children under 18, with head trauma being the leading cause of death in children below 5. A large but unknown number of traumatic injuries are non-accidental, i.e. inflicted. The lack of sensitivity and specificity required to diagnose Abusive Head Trauma (AHT) from radiological studies results in putting the children at risk of re-injury and death. Modern Deep Learning techniques can be utilized to detect Abusive Head Trauma using Computer Tomography (CT) scans. Training models using these techniques are only a part of building AI-driven Computer-Aided Diagnostic systems. There are challenges in deploying the models to make them highly available and scalable. The thesis models the domain of Abusive Head Trauma using Deep Learning techniques and builds an AI-driven System at scale using best Software Engineering Practices. It has been done in collaboration with Phoenix Children Hospital (PCH). The thesis breaks down AHT into sub-domains of Medical Knowledge, Data Collection, Data Pre-processing, Image Generation, Image Classification, Building APIs, Containers and Kubernetes. Data Collection and Pre-processing were done at PCH with the help of trauma researchers and radiologists. Experiments are run using Deep Learning models such as DCGAN (for Image Generation), Pretrained 2D and custom 3D CNN classifiers for the classification tasks. The trained models are exposed as APIs using the Flask web framework, contained using Docker and deployed on a Kubernetes cluster. The results are analyzed based on the accuracy of the models, the feasibility of their implementation as APIs and load testing the Kubernetes cluster. They suggest the need for Data Annotation at the Slice level for CT scans and an increase in the Data Collection process. Load Testing reveals the auto-scalability feature of the cluster to serve a high number of requests. / Dissertation/Thesis / Masters Thesis Software Engineering 2020

Dans le domaine médical, l'informatisation des professions de santé et le développement du dossier médical personnel (DMP) entraîne une progression rapide du volume d'information médicale numérique. Le besoin de convertir et de manipuler toute ces informations sous une forme structurée constitue un enjeu majeur. C'est le point de départ de la mise au point d'outils d'interrogation appropriés pour lesquels, les méthodes issues du traitement automatique du langage naturel (TALN) semblent bien adaptées. Les travaux de cette thèse s'inscrivent dans le domaine de l'analyse de documents médicaux et traitent de la problématique de la représentation de l'information biomédicale (en particulier du domaine radiologique) et de son accès. Nous proposons de construire une base de connaissance dédiée à la radiologie à l'intérieur d'une base de connaissance générale (réseau lexico-sémantique JeuxDeMots). Nous montrons l'intérêt de l'hypothèse de non séparation entre les différents types de connaissances dans le cadre d'une analyse de documents. Cette hypothèse est que l'utilisation de connaissances générales, en plus de celles de spécialités, permet d'améliorer significativement l'analyse de documents médicaux.Au niveau du réseau lexico-sémantique, l'ajout manuel et automatisé des méta-informations sur les annotations (informations fréquentielles, de pertinences, etc) est particulièrement utile. Ce réseau combine poids et annotations sur des relations typées entre des termes et des concepts ainsi qu'un mécanisme d'inférence dont l'objet est d'améliorer la qualité et la couverture du réseau. Nous décrivons comment à partir d'informations sémantiques présentes dans le réseau, il est possible de définir une augmentation des index bruts construits pour chaque comptes rendus afin d'améliorer la recherche documentaire. Nous présentons, ensuite, une méthode d'extraction de relations sémantiques entre des termes ou concepts. Cette extraction est réalisée à l'aide de patrons linguistiques auxquels nous avons rajouté des contraintes sémantiques.Les résultats des évaluations montrent que l'hypothèse de non séparation entre les différents types de connaissances améliorent la pertinence de l'indexation. L'augmentation d'index permet une amélioration du rappel alors que les contraintes sémantiques améliorent la précision de l'extraction de relations. / In the medical field, the computerization of health professions and development of the personal medical file (DMP) results in a fast increase in the volume of medical digital information. The need to convert and manipulate all this information in a structured form is a major challenge. This is the starting point for the development of appropriate tools where the methods from the natural language processing (NLP) seem well suited.The work of this thesis are within the field of analysis of medical documents and address the issue of representation of biomedical information (especially the radiology area) and its access. We propose to build a knowledge base dedicated to radiology within a general knowledge base (lexical-semantic network JeuxDeMots). We show the interest of the hypothesis of no separation between different types of knowledge through a document analysis. This hypothesis is that the use of general knowledge, in addition to those specialties, significantly improves the analysis of medical documents.At the level of lexical-semantic network, manual and automated addition of meta information on annotations (frequency information, pertinence, etc.) is particularly useful. This network combines weight and annotations on typed relationships between terms and concepts as well as an inference mechanism which aims to improve quality and network coverage. We describe how from semantic information in the network, it is possible to define an increase in gross index built for each records to improve information retrieval. We present then a method of extracting semantic relationships between terms or concepts. This extraction is performed using lexical patterns to which we added semantic constraints.The results show that the hypothesis of no separation between different types of knowledge to improve the relevance of indexing. The index increase results in an improved return while semantic constraints improve the accuracy of the relationship extraction.

Breast cancer is the second fatal disease among cancers both in Canada and across the globe. However, in the case of early detection, it can raise the survival rate. Thus, researchers and scientists have been practicing to develop Computer-Aided Diagnosis (CAD)x systems. Traditional CAD systems depend on manual feature extraction, which has provided radiologists with poor detection and diagnosis tools. However, recently the application of Convolutional Neural Networks (CNN)s as one of the most impressive deep learning-based methods and one of its interesting techniques, Transfer Learning, has revolutionized the performance and development of these systems. In medical diagnosis, one issue is distinguishing between breast mass lesions and calcifications (little deposits of calcium). This work offers a solution using transfer learning and ensemble learning (majority voting) at the first stage and later replacing the voting strategy with soft voting. Also, regardless of the abnormality's type (mass or calcification), the severeness of the abnormality plays a key role. Nevertheless, in this study, we went further and made an effort to create a (CAD)x pathology diagnosis system. More specifically, after comparing multi-classification results with a two-staged abnormality diagnosis system, we propose the two-staged binary classifier as our final model. Thus, we offer a novel breast cancer diagnosis system using a wide range of pre-trained models in this study. To the best of our knowledge, we are the first who integrate the application of a wide range of state-of-the-art pre-trained models, particularly including EfficientNet for the transfer learning part, and subsequently, employ ensemble learning. With the application of pre-trained CNN-based models or transfer learning, we are able to overcome the lack of large-size datasets. Moreover, with the EfficientNet family offering better results with fewer parameters, we achieved promising results in terms of accuracy and AUC-score, and later ensemble learning was applied to provide robustness for the network. After performing 10-fold cross-validation, our experiments yielded promising results; while constructing the breast abnormality classifier 0.96 ± 0.03 and 0.96 for accuracy and AUC-score, respectively. Similarly, it resulted in 0.85 ± 0.08 for accuracy and 0.81 for AUC-score when constructing pathology diagnosis.

Mammography is used to aid early detection and diagnosis systems. It takes an x-ray image of the breast and can provide a second opinion for radiologists. The earlier detection is made, the better treatment works. Digital mammograms are dealt with by Computer Aided Diagnosis (CAD) systems that can detect and analyze abnormalities in a mammogram. The purpose of this study is to investigate how to categories cropped regions of interest (ROI) from digital mammogram images into two classes; normal and abnormal regions (which contain microcalcifications). The work proposed in this thesis is divided into three stages to provide a concept system for classification between normal and abnormal cases. The first stage is the Segmentation Process, which applies thresholding filters to separate the abnormal objects (foreground) from the breast tissue (background). Moreover, this study has been carried out on mammogram images and mainly on cropped ROI images from different sizes that represent individual microcalcification and ROI that represent a cluster of microcalcifications. The second stage in this thesis is feature extraction. This stage makes use of the segmented ROI images to extract characteristic features that would help in identifying regions of interest. The wavelet transform has been utilized for this process as it provides a variety of features that could be examined in future studies. The third and final stage is classification, where machine learning is applied to be able to distinguish between normal ROI images and ROI images that may contain microcalcifications. The result indicated was that by combining wavelet transform and SVM we can distinguish between regions with normal breast tissue and regions that include microcalcifications.

No description available.

No description available.

In this study, the impact of multiple image preprocessing methods on Convolutional Neural Networks (CNN) was studied. Metrics such as accuracy, precision, recall and F1-score (Hossin et al. 2011) were evaluated. Specifically, this study is geared towards malaria classification using the data set made available by the U.S. National Library of Medicine (Malaria Datasets n.d.). This data set contains images of thin blood smears, where uninfected and parasitized blood cells have been segmented. In the study, 3 CNN models were proposed for the parasitization classification task. Each model was trained on the original data set and 4 preprocessed data sets. The preprocessing methods used to create the 4 data sets were grayscale, normalization, histogram equalization and contrast limited adaptive histogram equalization (CLAHE). The impact of CLAHE preprocessing yielded a 1.46% (model 1) and 0.61% (model 2) improvement over the original data set, in terms of F1-score. One model (model 3) provided inconclusive results. The results show that CNN’s can be used for parasitization classification, but the impact of preprocessing is limited. / I denna studie studerades effekten av flera bildförbehandlingsmetoder på Convolutional Neural Networks (CNN). Mätvärden såsom accuracy, precision, recall och F1-score (Hossin et al. 2011) utvärderades. Specifikt är denna studie inriktad på malariaklassificering med hjälp av ett dataset som tillhandahålls av U.S. National Library of Medicine (Malaria Datasets n.d.). Detta dataset innehåller bilder av tunna blodutstryk, med segmenterade oinfekterade och parasiterade blodceller. I denna studie föreslogs 3 CNN-modeller för parasiteringsklassificeringen. Varje modell tränades på det ursprungliga datasetet och 4 förbehandlade dataset. De förbehandlingsmetoder som användes för att skapa de 4 dataseten var gråskala, normalisering, histogramutjämning och kontrastbegränsad adaptiv histogramutjämning (CLAHE). Effekten av CLAHE-förbehandlingen gav en förbättring av 1.46% (modell 1) och 0.61% (modell 2) jämfört med det ursprungliga datasetet, vad gäller F1-score. En modell (modell 3) gav inget resultat. Resultaten visar att CNN:er kan användas för parasiteringsklassificering, men effekten av förbehandling är begränsad.