IVAN MACIA

Ph. D.

Fundación Centro de Tecnologías de Interacción Visual y Comunicaciones - Vicomtech

Director of Digital Health and Biomedical Technologies, PI

NLST (Learn more about this study)

NLST-1149

Oct 31, 2023

Advancing Lung Cancer Detection with AI: Screening Models and Risk Factor Analysis

Lung cancer is a significant global public health concern, with an estimated 2.2 million new diagnoses and up to 1.8 million deaths reported worldwide in 2020. The five-year overall survival rate, when age-standardized, remains consistently low, ranging between 10% and 20% in most countries. Despite the existence of screening programs designed to detect lung cancer at early stages, timely and effective diagnosis remains a complex challenge.

Our research project is based on the hypothesis that a comprehensive analysis of multimodal data with advanced AI algorithms, including Machine Learning and Deep Learning, can
enhance our understanding of the etiology of lung cancer and could provide better algorithms for high risk individual selection for risk-based screening. This could lead to the following developments leading to a better understanding and screening of lung cancer. Firstly, the development of more effective screening programs that can refine patient stratification into high and low-risk categories, leading to improved outcomes for individuals at risk. Secondly, a more thorough identification and understanding of risk factors contributing to the development of lung cancer and their impact. Thirdly, the development of models that can predict more accurately the risk of lung cancer based on risk factors, clinical data and CT scan images. And lastly, the image-based characterization of the tumor microenvironment identifying patterns of cancer progression.

To achieve our research goals, we propose a 3-phased approach:
- In the first phase, we will focus on developing advanced lung cancer screening methods and gaining insights into risk factors associated to its development. This will involve implementing AI
models based on cutting-edge algorithms and benchmarking them against state-of-the-art screening models such as PLCOm2012. Subsequently, we will identify the best-performing
model and employ feature selection methodology to understand the significance of the key risk factors contributing to its performance. Here, our goal is to explore the relationships and relative importance among established risk factors as well as potentially discover new ones.
- In the second phase, our efforts will center on developing models with a dual purpose: enhanced lung cancer case detection in screening CTs and estimating the short-to-medium-term risk of
developing the disease, combining structured data and imaging data to derive advanced risk prediction models based on multimodal data. Finally, we will assess its performance on an independent cohort from the LUCIA initiative, an Horizon Europe project from the Cancer Mission, aimed at improved our understanding of lung cancer and implementation of effective screening programmes.
- In the last phase, our focus shifts to understanding the tumor microenvironment in relation to its progression. We will begin by fine-tuning image segmentation algorithms to accurately
identify relevant substructures within the tumor microenvironment. This process includes training and testing our models using the TCGA. Our primary goal is to develop an image
segmentation algorithm capable of distinguishing between histological substructures. Subsequently, we will apply this algorithm to lung cancer samples from NLST with the aim of
identifying specific segmentation patterns correlated with tumor subtypes and prognosis.