Advanced Deep Learning for Early Detection and Prognostication in Lung Cancer
Principal Investigator
Name
Zeng Zeng
Degrees
Ph.D.
Institution
Institute for Infocomm Research
Position Title
Scientist
Email
zengz@i2r.a-star.edu.sg
Project Information
Study
NLST
Project ID
NLST-323
Title
Advanced Deep Learning for Early Detection and Prognostication in Lung Cancer
Summary

Lung cancer contributes 20% of cancer related mortalities. In particular, the incidence of lung cancer in South and East Asia is growing at an alarming rate. Significant research and resources are being invested to better diagnose and manage this disease. However, the morbidity and mortality associated with this disease remain significant. Techniques to accurately diagnose and prognosticate lung cancer are limited. Prevailing challenges include high rate of false positives with low dose CT scans, as well as the inability to quantitatively evaluate progress and prognosticate outcomes. Therefore, there is a pressing need to develop techniques that can address these limitations.

Such an endeavor would require data from comprehensive studies that collate clinical and demographic information pertaining to lung cancer diagnosis and progression over time. As these studies are limited in the Asian context, the NLST dataset offers a unique resource that we can apply and validate our methods with.

In this project, we aim to use the NLST dataset to develop and validate novel deep learning approaches for early detection and prognostication in lung cancer. We will use a combination of the time-lapse CT scans, pathological images, medical record and demographical information to train our models. As these datasets are multimodal and contain features that evolve over time, we propose to apply deep learning methodologies that can deal with heterogeneous data and adapt to time-varying features. We posit that accounting for the composite information within the NLST data as the patient’s state evolves over time will enable earlier detection of cancerous nodules and better evaluation of disease progression. The proposed work will ultimately contribute to improved clinical decision support for lung cancer management.
Aims

1. Develop deep learning methods to detect cancerous nodules with low dose time-lapse lung CT scans.
2. Develop advanced deep learning approaches to integrate multimodal clinical data (images, medical record and demographical information) for improved prognostication in lung cancer.
Collaborators

Dr. Lee Hwee Kuan, Bioinformatics Institute, A*STAR, Singapore
Dr. Daniel Tan Shao Weng, National Cancer Center, Singapore
Dr. Babar Nazir, National Cancer Center, Singapore
Dr. Vijay Ramaseshan Chandrasekhar, Institute for Infocomm Research, A*STAR, Singapore
Dr. Pavitra Krishnaswamy, Institute for Infocomm Research, A*STAR, Singapore
Dr. Nanying Liang, Institute for Infocomm Research, A*STAR, Singapore