Biomedical Image Analytics: Automated Lung Cancer Diagnosis

Abstract Biomedical informatics as an emerging field has been fascinating talents from artificial intelligence and machine learning for its unique opportunities and challenges. Fast-growing biomedical and healthcare data have encompassed multiple scales ranging from molecules, cells, and individuals to populations, and have connected various entities in healthcare systems such as providers, pharma, and payers with increasing bandwidth, depth, and resolution. These data are becoming an enabling resource to harness for scientific knowledge discovery and clinical decision making. Meanwhile, the sheer volume and complexity of the data present major barriers toward their translation into effective clinical actions. In particular, biomedical data often feature large volumes, high dimensions, imbalance between classes, heterogeneous sources, noises, incompleteness, and rich contexts, which challenges the direct and immediate success of existing machine learning and optimization methods. For instance, deep learning methods have made notable advances for biomedical informatics needs, especially in processing brain-imaging data and making neuroscience discovery, although their utilities to more types of data in more biomedical informatics use-cases still awaits further assessment and development. Therefore, there is a compelling demand for novel algorithms, including machine learning, data mining and optimization, that specifically tackle the unique challenges associated with biomedical and healthcare data and allow decision-makers and stakeholders to better interpret and exploit the data [1] , [2] , [3] , [4] , [5] . One of the challenges of using machine learning techniques with medical data is the frequent dearth of source image data for training. In this chapter, we introduce automated lung cancer diagnosis as a representative example, where nodule images need to be classified as suspicious or benign. In this work we propose an automatic synthetic lung nodule image generator. Our 3D shape generator is designed to augment the variety of 3D images. Our proposed system takes root in autoencoder techniques, and we provide extensive experimental characterization that demonstrates its ability to produce quality synthetic images.