Integrative analysis to predict lung cancer patient outcome using NLST dataset
Lung cancer is the leading cause of death from cancer for both men and women in the United States. Better prediction of lung cancer patient prognosis and response to therapy will greatly facilitate treatment planning and thereby improve patient outcomes. Reliable computational methods to predict clinical outcomes using tumor tissue slides, which are available in standard clinical care, will have an immediate impact on lung cancer patient care.
Histopathological classifications of lung cancer have been one of the major criteria guiding lung cancer treatment, and recent studies show that more refined pathological classification of non-small-cell lung cancer can lead to better treatment decisions. However, this type of pathological grading system requires extensive information processing by a human brain to interpret highly-complex pathology images, which is time-consuming, subjective, and generates considerable inter- and intra-observer variation. Tumor tissue slide scanning is becoming a routine clinical procedure and can produce massive digital pathology images that capture histological details in high resolution. Reliable computational methods to predict clinical outcomes using such tumor pathology slides will have an immediate impact on patient care in lung cancer.
The overall goal of this proposed study is to develop pathology image-based outcome prediction models in lung cancer for both early detection and clinical outcomes (such as prognosis and response to therapy) using the X-ray, CT and pathology image, and clinical data from NLST study. The specific aims are as follows:
Aim 1. Develop algorithms for pathology image analysis in lung cancer. In this aim, we will develop computational approaches for pathology imaging data preprocessing and analysis. Furthermore, we will develop advanced deep learning algorithms to classify cell types for tumor pathology images. The accuracy of cell type classification will be validated visually by pathologists and experimentally using protein assays.
Aim 2. Develop clinical outcome prediction models from pathology imaging features. The spatial distributions of different types of cells (tumor cells, stromal cells, lymphocytes) could reveal a cancer cell’s growth pattern, its relationships with the tumor microenvironment and the immune response of the body. In this Aim, we will develop novel mathematical methods to effectively characterize the spatial patterns and interactions among different types of cells and their nearby alveoli, bronchioles and blood vessels. We will then use these measures together with the multi-level image features extracted from Aim 1 to develop prediction models for both patient prognosis and response to therapy.
Aim 3. Integrate X-ray, CT and pathology image, and clinical data from NLST study for lung cancer early detection and clinical outcome prediction. We will develop a large-scale multi-view learning algorithm to integrate X-ray, CT and pathology image, and clinical data from NLST study model for lung cancer early detection and clinical outcome prediction. We will also compare the prediction performance of using each individual type of data to evaluate the clinical utility of different types of data.