Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Daniel Rückert
Degrees
MSc, PhD
Institution
Technical University of Munich (TUM)
Position Title
Professor of Artificial Intelligence in Healthcare and Medicine
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-1075
Initial CDAS Request Approval
Jun 2, 2023
Title
Multimodal AI Algorithms for Diagnosis and Prognosis in Lung Cancer
Summary
I. Introduction

Lung Cancer (LC) remains a significant global health issue with high mortality rates, primarily due to late-stage diagnosis. This proposal seeks to optimize the diagnostic and prognostic decision-making process of LC through Artificial Intelligence (AI) and Machine Learning (ML) methodologies. Specifically, we intend to utilize the rich clinical and imaging data derived from the National Lung Screening Trial (NLST).

II. Objectives

Our project primarily focuses on the following areas:

Personalized Diagnosis: The objective is to enhance the diagnostic process by employing AI/ML algorithms to analyze low-dose Computed Tomography (CT) images with anomalies.

Personalized Prognosis: Our intent is to develop advanced AI/ML algorithms trained on multimodal data that enable predictive modeling of Progression Free Survival (PFS) and Overall Survival (OS) for individual patients.

III. Methodology

Our methodology involves a thorough exploratory data analysis as a prerequisite to assess the quality and suitability of data from the NLST. Following this, we plan to optimize an existing AI/ML algorithm, which employs radiomics features from CT scans, for its application in the LC patient population.

In the subsequent phase, we aim to enhance the performance of the algorithm by integrating advanced deep learning-based features. These additional features are expected to augment the diagnostic accuracy of the AI/ML model.

For personalized prognosis, our approach involves training AI/ML models on multimodal data to predict PFS and OS for individual patients. The performance of these models will be rigorously evaluated using robust statistical measures such as the concordance index, Brier score, and time-dependent Area Under the Curve (AUC).

IV. Expected Outcomes

Our project is designed to significantly advance LC diagnosis and prognosis, thus paving the way for more accurate and personalized patient care. The optimization and advancement of AI/ML algorithms for the interpretation of CT images can revolutionize LC diagnosis, potentially leading to better patient outcomes. Moreover, the development of personalized prognosis models can provide crucial insights into patient-specific survival rates, facilitating more effective treatment planning.

V. Conclusion

By leveraging the transformative capabilities of AI and ML, our project aims to make significant strides in the field of LC diagnosis and prognosis. Through more accurate diagnostic tools and personalized prognostic models, we anticipate a substantial positive impact on patient care, potentially reducing mortality and enhancing the quality of life for individuals afflicted with LC.
Aims

Better Patient Care: Through personalized diagnosis and prognosis, improve patient management strategies and treatment plans, enhancing overall patient outcomes.

Early Cancer Detection: Leverage AI and ML techniques to better detect lung cancer in its early stages from low-dose CT images, increasing the chances of successful treatment.

Better Prevention: Utilize the insights derived from the AI/ML models to identify potential risk factors, promoting better preventive measures.

Improving Prognostic Predictions: Develop robust algorithms to predict progression-free survival (PFS) and overall survival (OS) for individual patients, facilitating more informed decision-making regarding patient care.

Augmentation of Clinical Decision Making: Enable physicians to make more accurate and timely decisions by providing them with comprehensive and personalized patient data.

Enhancement of Research Capabilities: Foster better understanding of lung cancer, its progression, and response to treatment, thereby catalyzing further research in the field.

Data-Driven Health Policy: Influence health policy decisions and guidelines through the generation of comprehensive and reliable data on lung cancer detection, diagnosis, and prognosis.

Scalability and Transferability: Create algorithms and models that can be easily adapted for other disease areas, promoting broader application of our methods and findings.

Collaborators

Prof. Dr. Daniel Rückert, Institute of AI and Informatics in Medicine, Technical University of Munich