Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Arthus Carpentier
Institution
École Polytechnique
Position Title
Student
Email
About this CDAS Project
Study
PLCO (Learn more about this study)
Project ID
PLCO-1647
Initial CDAS Request Approval
Aug 15, 2024
Title
Advanced Neural network development for Early Detection and Localization of Cancers Through Proteomic Biomarkers.
Summary
Early detection of cancer is widely recognized as the most effective strategy in reducing cancer-related mortality and improving patient outcomes. At stages I and II, therapeutic interventions such as immunotherapy have demonstrated significant efficacy against minimal tumor burdens, rendering cancer a highly treatable disease with favorable response rates and tolerability profiles.

Currently, the majority of cancers are diagnosed through imaging techniques, which typically detect tumors when they reach a size on the order of millimeters. However, this approach has inherent limitations in sensitivity and scalability. In contrast, early screening via blood-based assays, capable of detecting tumors as early as the micron level, offers a more scalable and generalizable solution for early diagnosis. Nevertheless, the heterogeneity and sheer volume of circulating biomarkers and studies present a significant challenge for clinical interpretation. The complexity arises from the need to analyze and integrate a wide range of biomarkers, each potentially indicative of different cancer types or stages, which can overwhelm traditional diagnostic frameworks.
Aims

This research project proposes the development of advanced convolutional neural networks (CNNs) to systematically identify and interpret weak signals and emergent patterns of cancer development from comprehensive proteomic datasets. By leveraging deep learning algorithms, our goal is to enhance the specificity and sensitivity of early cancer detection, thereby enabling more precise and actionable clinical decision-making.

1. Cancer Localization Model: Given that many circulating proteins are common across multiple cancer types (e.g., CEA, CA 15-3), we will prioritize the development of a robust computational model designed to accurately localize the origin of the detected cancer. This model will integrate multi-modal data, including proteomic, genomic, and clinical features, to distinguish between different tumor origins with high precision.

2. Cancer Staging clustering: With a focus on improving patient management and optimizing therapeutic interventions, we will also develop a predictive model aimed at determining the stage of cancer progression in individual patients. This model will utilize a combination of clinical data, biomarker profiles, and imaging results to assess the extent of disease advancement. By predicting the stage of cancer, the model will support clinicians in stratifying patients more effectively, ensuring timely and appropriate care pathways, particularly for those presenting with advanced-stage cancers.

3. Cancer Prediction and Risk Factor Analysis - Longitudinal Analysis : In line with the goal of improving patient management, the strength of the PLCO study lies in its longitudinal data, which not only facilitates early cancer detection but also enables the development of predictive and preventive models that can anticipate cancer development several years before diagnosis. These models will analyze medical decision patterns that may contribute to cancer onset. Specifically, the long-term evolution of biomarkers such as CA-125 or PSA will be monitored to assess their correlation with the eventual development of cancer, few years later. Additionally, a comprehensive medical questionnaire will be developed to account for all risk factors and medical decisions that may increase the likelihood of cancer development, aiming to provide a proactive approach to patient care.

Collaborators

Alexandre Carpentier, Chief of Neurosurgery department at University Hospitals Pitié Salpêtrière