The adoption of Deep neural networks for lung cancer screening has been shown to improve detection of malignant nodules in sequential scans and to reduce the screening time1,2. However, DNNs fail to properly classify images when applied only to single low-dose computed tomography scans on indeterminate pulmonary nodules (4mm - 20mm in diameter). In addition, the limited size of most medical data sets utilized for deep learning leads to network overfitting and poor performance, making it difficult to translate to clinical settings3,4. The limited size of most medical dataset means that DNNs have difficulty identifying and evaluating features of interest and fail to generalize to novel data. However, guiding a neural network toward biological features that are known to be pathophysiologically relevant may improve both classification accuracy and generalizability5,6. For example, idiopathic pulmonary fibrosis and emphysema are both associated with increased lung inflammation and are considered pre-malignant conditions7-9. The aim of this work is to determine the contribution of biologically relevant features associated with increased risk of malignancy and embed them into deep learning methodologies to evaluate their contributions toward classification and model generalizability. Relevant biological features are identified through three methodologies: least absolute shrinkage and selection operator (LASSO), support vector machine (SVM), and random forest (RF). Using these quantitative features as the basis for classification, we increase the emphasis of specific tumor features through kernel manipulations using discrete wavelet decomposition and evaluate whether DNNs place emphasis on these select features for nodule classification. Results suggest that quantitative parenchymal features carry significant classification information across all machine learning methodologies applied. Using a combination of parenchymal features together with tumor specific features significantly improves classification performance of these methodologies compared to using only tumor specific features. Furthermore, DNNs extract abstract features that resemble these biological features when evaluating attention maps of the network. Features capturing nodule maximal diameter alongside textural and morphological features appear to drive nodule classification. The use of discrete wavelet decomposition to embed simplified features into CNNs improves classification accuracy of the model and reduces training time. This demonstrates that guiding a DNN toward select features can improve its performance while minimizing overfitting. The findings suggest that known pathophysiologically relevant features can be encoded into DNNs to improve network classification and generalizability to novel data. Furthermore, evaluating models based on misclassified nodules provides avenues to identify over-emphasized features in the network and correct them through image preprocessing. Overall, this body of work addresses several challenges present in the application of DNNs for early nodule detection. The performance of the generated models for single-shot classification of indeterminate pulmonary nodules shows promise for deployment as clinician co-pilots. Further studies on risk-benefits of these models in clinical settings are necessary to ensure proper performance prior to translation.

Low-dose computed tomography (CT) screening has been widely used to detect and diagnose early stage lung cancer. Clinical trials have shown that low-dose CT reduced lung cancer mortality by 20% relative to plain chest radiography; however, challenges exist in current low-dose CT screening programs including high over-diagnosis rates, high cost and increased radiation exposure. This dissertation attempts to overcome these challenges by developing machine and deep learning models for automated lung cancer diagnosis and disease progression estimation. A novel lung segmentation approach was first developed using a bidirectional chain code method and machine learning framework. This method is designed to include the lung nodules attached to lung wall while minimizing over-segmentation error. Second, a hybrid ensemble convolutional neural network has been developed to classify lung nodule vs. non-nodule objects. The ensemble model combines the VGG, residual and densely connected module designs to improve the model classification robustness for external datasets collected with different acquisition parameters. Third, a hierarchical semantic convolutional neural network (HSCNN) has been described to classify lung nodule malignancy. Semantic characteristic features, predicted in parallel with the malignancy for each nodule, enable the interpretation of the model and improvement of malignancy prediction. Finally, a Bayesian framework combined with a continuous-time Markov model was developed to estimate the multi-state disease progression of lung cancer. The resulting model estimates individual lung cancer state transition information, providing the basis for personalized screening recommendations. Extensive experiments and results have proved the effectiveness of these methods paving the way to optimize and improve the effectiveness of existing low-dose CT screening programs.

Chest X-ray imaging has become increasingly crucial for diagnosing various medical conditions, including pneumonia, lung cancer, and heart diseases. Despite the growing number of chest X-ray images, their interpretation remains a manual and time-consuming process, often leading to radiologist burnout and delays in diagnosis. The integration of domain knowledge and deep learning techniques has the potential to improve diagnosis, classification, and localization of abnormalities in chest X-rays, while also addressing the challenge of model interpretability. This work proposes a series of novel methods combining radiomics features and deep learning techniques for chest X-ray diagnosis, classification, and localization. We first introduce a framework leveraging radiomics features and contrastive learning for pneumonia detection, achieving superior performance and interpretability. The second method, ChexRadiNet, utilizes radiomics features and a lightweight triplet-attention mechanism for enhanced abnormality classification performance. In addition, we present a semi-supervised knowledge-augmented contrastive learning framework that seamlessly integrates radiomic features as a knowledge augmentation for disease classification and localization. This approach leverages Grad-CAM to highlight crucial abnormal regions, extracting radiomic features that act as positive samples for image features generated from the same chest X-ray. Consequently, this framework creates a feedback loop, enabling image and radiomic features to mutually reinforce each other, resulting in robust and interpretable knowledge-augmented representations. The Radiomics-Guided Transformer (RGT) fuses global image information with local radiomics-guided auxiliary information for accurate cardiopulmonary pathology localization and classification without bounding box annotations. Experimental results on public datasets such as NIH ChestX-ray, CheXpert, MIMIC-CXR, and the RSNA Pneumonia Detection Challenge demonstrate the effectiveness of our proposed methods, consistently outperforming state-of-the-art models in chest X-ray diagnosis, classification, and localization tasks. By bridging the gap between traditional radiomics and deep learning approaches, this work aims to advance the field of medical image analysis and facilitate more efficient and accurate diagnoses in clinical practice

Lung cancer remains the leading cause of cancer-related deaths worldwide. Early diagnosis can improve the effectiveness of treatment and increase a patient's chances of survival. Thus, there is an urgent need for new technology to diagnose small, malignant lung nodules early as well as large nodules located away from large diameter airways because

Lung cancer has a high incidence and mortality rate. The five-year relative survival rate for all lung cancers is 18%. Due to the high mortality and incidence rate of lung cancer worldwide, early detection is essential. Low dose Computed Tomography (CT) is a commonly used technique for screening, diagnosis, and prognosis of non-small cell lung cancer (NSCLC). The National Lung Screening Trial (NLST) compared low-dose helical computed tomography (LDCT) and standard chest radiography (CXR) for three annual screens and reported a 20% relative reduction in lung cancer mortality for LDCT compared to CXR. As such, LDCT screening for lung cancer is an effective way of mitigating lung cancer mortality and is the only imaging option for those at high risk. Lung cancer screening for high-risk patients often detects a large number of indeterminate pulmonary nodules, of which only a subset will be identified as cancer. As such, reliable and reproducible biomarkers determining which indeterminate pulmonary nodules will be identified as cancer would have significant translational implications as a therapeutic method to enhance lung cancer screening for nodule detection. Radiomics is an approach to extract high-dimensional quantitative features from medical images, which can be used individually or merged with clinical data for predictive and diagnostic analysis. Quantitative radiomics features (size, shape, and texture) extracted from lung CT scans have been shown to predict cancer incidence and prognosis. Deep learning is an emerging machine learning approach, which has been applied to the classification and analysis of various cancers and tumors. To generate generic features (blobs, edges, etc.) from an image, different convolutional kernels are applied over the input image, and then those generic feature-based images are passed through some fully connected neural layers. This category of the neural network is called a convolutional neural network (CNN), which has achieved high accuracy on image data. With the advancement of deep learning and convolutional neural networks (CNNs), deep features can be utilized to analyze lung CTs for prognosis prediction and diagnosis. In this dissertation, deep learning-based approaches were presented for lung nodule malignancy prediction. A subset of cases from the NLST was chosen as a dataset in our study. We experimented with three different pre-trained CNNs for extracting deep features and used five different classifiers. Experiments were also conducted with deep features from different color channels of a pre-trained CNN. Selected deep features were combined with radiomics features. Three CNNs were designed and trained. Combinations of features from pre-trained, CNNs trained on NLST data, and classical radiomics were used to build classifiers. The best accuracy (76.79%) was obtained using feature combinations. An area under the receiver operating characteristic curve of 0.87 was obtained using a CNN trained on an augmented NLST data cohort. After that, each of the three CNNs was trained using seven different seeds to create the initial weights. These enabled variability in the CNN models, which were combined to generate a robust, more accurate ensemble model. Augmenting images using only rotation and flipping and training with images from T0 yielded the best accuracy to predict lung cancer incidence at T2 from a separate test cohort (Accuracy = 90.29%; AUC = 0.96) based on an ensemble 21 models. From this research, five conclusions were obtained, which will be utilized in future research. First, we proposed a simple and effective CNN architecture with a small number of parameters useful for smaller (medical) datasets. Second, we showed features obtained using transfer learning with all the channels of a pre-trained CNN performed better than features extracted using any single channel and we also constructed a new feature set by fusing quantitative features with deep features, which in turn enhanced classification performance. Third, ensemble learning with deep neural networks was a compelling approach that accurately predicted lung cancer incidence at the second screening after the baseline screen, mostly two years later. Fourth, we proposed a method for deep features to have a recognizable definition via semantic or quantitative features. Fifth, deep features were dependent on the scanner parameters, and the dependency was removed using pixel size based normalization.

Convolutional Neural Networks for Medical Applications consists of research investigated by the author, containing state-of-the-art knowledge, authored by Dr Teoh Teik Toe, in applying Convolutional Neural Networks (CNNs) to the medical imagery domain. This book will expose researchers to various applications and techniques applied with deep learning on medical images, as well as unique techniques to enhance the performance of these networks.Through the various chapters and topics covered, this book provides knowledge about the fundamentals of deep learning to a common reader while allowing a research scholar to identify some futuristic problem areas. The topics covered include brain tumor classification, pneumonia image classification, white blood cell classification, skin cancer classification and diabetic retinopathy detection. The first chapter will begin by introducing various topics used in training CNNs to help readers with common concepts covered across the book. Each chapter begins by providing information about the disease, its implications to the affected and how the use of CNNs can help to tackle issues faced in healthcare. Readers would be exposed to various performance enhancement techniques, which have been tried and tested successfully, such as specific data augmentations and image processing techniques utilized to improve the accuracy of the models.