Chest Radiograph Interpretation with Deep Learning

Chest radiographs are one of the most commonly used diagnostic modalities in healthcare due to their effectiveness in detecting conditions related to the thoracic region. However, the increasing demand for radiological services, coupled with a shortage of radiologists and the potential for diagnostic errors, demands innovative solutions. The integration of artificial intelligence (AI) and deep learning techniques offers a promising approach to support radiology professionals and enhance the diagnostic accuracy and efficiency of the diagnostic process. This thesis addresses the challenge of improving the performance of deep learning models for multi-label classification of chest radiographs using the CheXpert dataset. It focuses on both model-centric and data-centric methods, specifically techniques such as Gaussian Mixture Models (GMM) for relabeling uncertain labels, multi-scale template matching for focused learning, a custom pooling layer for better feature extraction, and Sequential Multi-Label Enrichment (SMLE) for improved detection of coexisting conditions. The primary objective is to enhance the reliability and performance of AI models in chest radiograph interpretation by developing new model- and data-centric methods. This research aims to improve uncertain label handling, overall classification performance, and the detection of multiple coexisting conditions. The study employs convolutional neural networks (CNNs), DenseNet121, and Swin Transformers to investigate the effects of excluding versus relabeling uncertain labels. It also evaluates the efficacy of a custom pooling layer for classification performance and the effectiveness of theSMLEDenseNet model on co-existing conditions. Significant performance improvements are demonstrated when uncertain labels are appropriately handled using GMM. The custom pooling layer significantly enhances the classification performance of the model, and the SMLE DenseNet model outperforms the baseline DenseNet121 in detecting co-existing conditions. Finally, the study also examines radiologist’s shift in perception towards the integration of AI in clinical practice through pre- and post-presentation questionnaires. The findings indicate a generally positive attitude shift towards AI. Despite concerns regarding the deskilling of new radiology professionals, radiologists recognize the potential of AI models to enhance the efficiency of radiology departments and patient care. This research contributes insights into the practical implementation of AI in medical imaging. It shows that advanced techniques can significantly improve the diagnostic performance and efficiency of AI models. The findings also emphasize the importance of addressing radiologist’s concerns to ensure the successful integration of AI into clinical practice.