Thesis (MSc. (Faculty of Health Sciences))--University of Pretoria, 2007. / Summary in English and Afrikaans. Includes bibliographical references.

Estimates of left atrial size in patients with suspected cardiac disease play an important role in diagnostic medicine. Left atrial size is used in predicting prognosis and events, as well as treatment decisions. Two methods are commonly used to estimate left atrial size: chest radiography and cardiac ultrasound. This study aims to determine the test characteristics of chest radiography and compare the use of radiographs to cardiac ultrasound (the gold standard test). Data from patients older than 18 years admitted to Pretoria Academic Hospital during 2000-2003 who had both chest X-rays and cardiac ultrasound were included in this cross-sectional, retrospective analysis. Chest X-rays were classified into three quality classes, and the sub-carinal angle (SCA) and sub-angle distance (SAD) were measured twice in all available X-rays by two observers. Intra- and interobserver variability (3 methods) as well as the predictive value of the SCA and SAD measurements were determined using logistic regression (with left atrial size determined by ultrasound as comparator). P-values < 0.05 were regarded as statistically significant for all comparisons. Data for 159 patients were available (154 cardiac ultrasounds and 178 chest radiographs). Intraobserver variability regarding chest X-ray measurements was low with almost perfect concordance (P=0.000). Interobserver variability was higher for supine X-rays. Using logistic regression, a linear model was identified which was statistically significant only for erect X-rays. While goodness-of-fit analysis showed that the model fits the data, performance characteristics were poor, with high sensitivity and low specificity, and an area under the ROC curve of 0.62-0.63, depending on type of X-ray and measurement (SCA or SAD). Linearity in the logit of the dependent variable was assessed, and found to be present at the extremes of carinal angle measurements for the supine X-ray data and in the first three quartiles for erect X-ray data. A non-linear model determined by fractional polynomial analysis did not perform significantly better than the original linear model. Cut-off values for the SCA of 72o and 84o (erect and supine X-rays, respectively) were found to give the best compromise between sensitivity and specificity. The corresponding cut-off values for SAD were 24.1mm and 26.9mm. Assessment of either SCA or SAD to determine left atrial size is equivalent and repeatable, both within the same observer, and between two observers (less so for supine X-rays). While this measure is precise, it was found not to be very accurate. Therefore, chest X-rays are not reliable in predicting left atrial enlargement. / Dissertation (MSc (Clinical Epidemiology))--University of Pretoria, 2007. / Clinical Epidemiology / unrestricted

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

Das ARDS gilt, trotz verschiedenster Möglichkeiten der intensivmedizinischen Therapie, immer noch als die schwerste Form einer Lungenparenchymverletzung mit einer hohen Letalität. Zu Beginn der Erkrankung zeigt sich eine große Diskrepanz zwischen zunehmender Hypoxie und blandem Röntgen-Thorax-Befund. Die Röntgen-Thorax-Aufnahme bildet somit einen wichtigen diagnostischen Bestandteil. Die Kriterien einer einfachen Durchführung, guten Reproduzierbarkeit, hohen Aussagekraft und der Möglichkeit einer Verlaufsbeurteilung machen die Röntgen-Thorax-Liegendaufnahme zu einem wichtigen Bestandteil der intensivmedizinischen Diagnostik. Untersucht wurden retrospektiv 1575 Röntgen-Thorax-Aufnahmen von 33 Patienten mit dem Krankheitsbild des ARDS. Das Patientenkollektiv setzte sich aus 14 Frauen und 19 Männern im Alter von 12 bis 63 Jahren zusammen. Die Überlebensrate des untersuchten Patientenkollektives betrug 87,9%. Pro Patient wurden durchschnittlich 48 Röntgen-Thorax-Aufnahmen (zwischen 10 und 148 Aufnahmen) angefertigt. Die Aufnahmen wurden im Hinblick auf ihre Qualität, Strahlenexposition und ihre Übereinstimmung mit klinischen Parametern untersucht. Für die Beurteilung der Inter- und Intraobservervariabilität wurden verschiedenen Untersuchern 60 Röntgen-Thorax-Aufnahmen exemplarisch zur Bewertung vorgelegt. Aufgrund der schwierigen Aufnahmebedingungen bei Intensivpatienten sind verdrehte und verkippte Röntgen-Thorax-Aufnahmen nicht zu vermeiden. Trotz dieser Qualitätseinbußen ist ihr Informationsgehalt ein wichtiges Kriterium der intensivmedizinischen Diagnose und Therapie. Die Strahlenexposition der Röntgen-Thorax-Liegendaufnahmen und der daraus zu errechnende Lebenszeitverlust sind im Hinblick auf die Schwere und die hohe Letalität der Grunderkrankung als verschwindend gering zu betrachten. Die Röntgen-Scores nach Murray, Morel, Miniati, Rommelsheim und Ostendorf sind in der Diagnostik und Verlaufsbeurteilung des ARDS weit verbreitet. Sie sind in Handhabung und Gewichtung der Veränderungen jedoch sehr unterschiedlich. Zusammenhänge zwischen klinischen Parametern konnten für alle Scores, sowie für einen neuen Score beobachtet werden. Dabei ergaben sich für alle Scores ähnliche Beziehungen. Unter Berücksichtigung der Handhabung der einzelnen Röntgen-Scores, der Inter- und Intraobservervariabilität, sowie der Übereinstimmung mit klinischen Parametern sind der Röntgen-Score nach Rommelsheim, sowie der neue Röntgen-Score für den klinischen Alltag zu empfehlen. / Radiological Quantification of Chest X-Rays of the Lung Oedema of Patients with Adult Respiratory Distress Syndrome (ARDS) The ARDS is still regarded as the most serious form of lung parenchyma injury with a high lethality in spite of various possibilities of intensive care therapies. In the beginning of the illness a high discrepancy between an increasing hypoxia and a mostly inconspicouos chest X-ray result can be observed. Therefore, the chest X-ray forms an essential part of the diagnostic basis. It is characterized by simple implementation, a good reproducibility, a high meaningfulness and the possibility to judge the course of the illness. This makes the chest X-ray very valuable for the intensive care diagnosis. 1575 chest X-rays from 33 patients with ARDS symptoms were evaluated. The patients consisted of 14 women and 19 men between 12 and 63 years of age. The overall survival rate for all patients was 87.9%. An average of 48 (ranging from 10 to 148) chest X-rays were taken per patient. They were examined with regard to quality, radiation dose and correspondence to clinical variables. In order to judge the interobserver and intraobserver variability 60 chest X-rays were evaluated by different examiners. Due to the difficult conditions while taking the chest X-rays distorted and tilted chest X-rays cannot be avoided. Despite this loss of quality their content of information is an important criterion for the intensive care diagnosis and treatment. The loss of lifetime due to the radiation dose received can be ignored compared to the severeness and the high lethality of the basic illness. The chest X-ray scores according to Murray, Morel, Miniati, Rommelsheim and Ostendorf are commonly used for diagnostic purposes and to judge the course of the ARDS. However, their handling and their weightning of changes varies a lot. Correlations between all scores, a new score and the clinical variables were observed. All scores, including the new score, showed similar relations between the score ranking and the clinical variables. Considering the handling of the different chest X-ray scores, their interobserver and intraobserver variability and their correlation to clinical variables the chest X-ray score according to Rommelsheim and the new score can be recommended for daily use.

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal

This thesis addresses the extraction of medical knowledge from clinical text using deep learning techniques. In particular, the proposed methods focus on cancer clinical trial protocols and chest x-rays reports. The main results are a proof of concept of the capability of machine learning methods to discern which are regarded as inclusion or exclusion criteria in short free-text clinical notes, and a large scale chest x-ray image dataset labeled with radiological findings, diagnoses and anatomic locations. Clinical trials provide the evidence needed to determine the safety and effectiveness of new medical treatments. These trials are the basis employed for clinical practice guidelines and greatly assist clinicians in their daily practice when making decisions regarding treatment. However, the eligibility criteria used in oncology trials are too restrictive. Patients are often excluded on the basis of comorbidity, past or concomitant treatments and the fact they are over a certain age, and those patients that are selected do not, therefore, mimic clinical practice. This signifies that the results obtained in clinical trials cannot be extrapolated to patients if their clinical profiles were excluded from the clinical trial protocols. The efficacy and safety of new treatments for patients with these characteristics are not, therefore, defined. Given the clinical characteristics of particular patients, their type of cancer and the intended treatment, discovering whether or not they are represented in the corpus of available clinical trials requires the manual review of numerous eligibility criteria, which is impracticable for clinicians on a daily basis. In this thesis, a large medical corpora comprising all cancer clinical trials protocols in the last 18 years published by competent authorities was used to extract medical knowledge in order to help automatically learn patient’s eligibility in these trials. For this, a model is built to automatically predict whether short clinical statements were considered inclusion or exclusion criteria. A method based on deep neural networks is trained on a dataset of 6 million short free-texts to classify them between elegible or not elegible. For this, pretrained word embeddings were used as inputs in order to predict whether or not short free-text statements describing clinical information were considered eligible. The semantic reasoning of the word-embedding representations obtained was also analyzed, being able to identify equivalent treatments for a type of tumor in an analogy with the drugs used to treat other tumors. Results show that representation learning using deep neural networks can be successfully leveraged to extract the medical knowledge from clinical trial protocols and potentially assist practitioners when prescribing treatments. The second main task addressed in this thesis is related to knowledge extraction from medical reports associated with radiographs. Conventional radiology remains the most performed technique in radiodiagnosis services, with a percentage close to 75% (Radiología Médica, 2010). In particular, chest x-ray is the most common medical imaging exam with over 35 million taken every year in the US alone (Kamel et al., 2017). They allow for inexpensive screening of several pathologies including masses, pulmonary nodules, effusions, cardiac abnormalities and pneumothorax. For this task, all the chest-x rays that had been interpreted and reported by radiologists at the Hospital Universitario de San Juan (Alicante) from Jan 2009 to Dec 2017 were used to build a novel large-scale dataset in which each high-resolution radiograph is labeled with its corresponding metadata, radiological findings and pathologies. This dataset, named PadChest, includes more than 160,000 images obtained from 67,000 patients, covering six different position views and additional information on image acquisition and patient demography. The free text reports written in Spanish by radiologists were labeled with 174 different radiographic findings, 19 differential diagnoses and 104 anatomic locations organized as a hierarchical taxonomy and mapped onto standard Unified Medical Language System (UMLS) terminology. For this, a subset of the reports (a 27%) were manually annotated by trained physicians, whereas the remaining set was automatically labeled with deep supervised learning methods using attention mechanisms and fed with the text reports. The labels generated were then validated in an independent test set achieving a 0.93 Micro-F1 score. To the best of our knowledge, this is one of the largest public chest x-ray databases suitable for training supervised models concerning radiographs, and also the first to contain radiographic reports in Spanish. The PadChest dataset can be downloaded on request from http://bimcv.cipf.es/bimcv-projects/padchest/. PadChest is intended for training image classifiers based on deep learning techniques to extract medical knowledge from chest x-rays. It is essential that automatic radiology reporting methods could be integrated in a clinically validated manner in radiologists’ workflow in order to help specialists to improve their efficiency and enable safer and actionable reporting. Computer vision methods capable of identifying both the large spectrum of thoracic abnormalities (and also the normality) need to be trained on large-scale comprehensively labeled large-scale x-ray datasets such as PadChest. The development of these computer vision tools, once clinically validated, could serve to fulfill a broad range of unmet needs. Beyond implementing and obtaining results for both clinical trials and chest x-rays, this thesis studies the nature of the health data, the novelty of applying deep learning methods to obtain large-scale labeled medical datasets, and the relevance of its applications in medical research, which have contributed to its extramural diffusion and worldwide reach. This thesis describes this journey so that the reader is navigated across multiple disciplines, from engineering to medicine up to ethical considerations in artificial intelligence applied to medicine.