Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Wenchuan Zhang
Degrees
Ph.D student
Institution
Department of pathology of West China hospital,China
Position Title
student
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-1261
Initial CDAS Request Approval
May 29, 2024
Title
Developing an AI diagnostic model for lung cancer based on deep learning algorithms using histopathology images and imaging images
Summary
Lung cancer is the leading cause of cancer-related deaths worldwide, necessitating the development of more accurate diagnostic tools. This project aims to develop an AI-based diagnostic model for lung cancer by integrating deep learning algorithms with histopathology and imaging images. By combining these data sources, we aim to create a robust model capable of distinguishing between benign and malignant lesions, grading the severity of cancer, and potentially predicting patient prognosis.

Histopathology images offer detailed cellular-level information crucial for identifying cancerous changes, while imaging images (such as CT scans) provide broader anatomical context and functional information. The integration of these modalities will leverage the strengths of each, providing a comprehensive analysis that could improve diagnostic accuracy and early detection.

Our model will be trained and validated using a large, multi-institutional dataset to ensure generalizability and robustness. We will employ state-of-the-art deep learning techniques, including convolutional neural networks (CNNs) and advanced data augmentation strategies, to handle the complexity and variability of medical images. Furthermore, we will integrate clinical data to enhance the model's predictive power.

The successful development of this AI diagnostic model has the potential to revolutionize lung cancer diagnosis, providing clinicians with a powerful tool to support decision-making and ultimately improve patient care.
Aims

Aim 1: Develop a deep learning model for lung cancer diagnosis using histopathology images.
Objective: To create and train a convolutional neural network (CNN) that can accurately classify histopathology images as benign or malignant and grade the severity of lung cancer.
Methods: We will preprocess and augment a large dataset of histopathology images, ensuring a balanced representation of different cancer types and grades. The CNN architecture will be optimized through hyperparameter tuning and cross-validation.
Expected Outcomes: A highly accurate model that can distinguish between benign and malignant lung tissues and provide detailed grading information.

Aim 2: Develop a deep learning model for lung cancer diagnosis using imaging images.
Objective: To create and train a CNN that can analyze imaging images (e.g., CT scans) for the presence and characteristics of lung cancer.
Methods: We will use a large dataset of imaging images, applying advanced preprocessing techniques to enhance image quality and augment the data. The model will be trained to identify and characterize lung lesions.
Expected Outcomes: An effective model that can identify lung cancer lesions with high sensitivity and specificity, providing valuable information on lesion size, location, and characteristics.

Aim 3: Integrate histopathology and imaging data to develop a multimodal AI diagnostic model for lung cancer.
Objective: To create a comprehensive diagnostic model that combines histopathology and imaging data to improve diagnostic accuracy and prognostic predictions.
Methods: We will develop a fusion network that integrates the outputs of the histopathology and imaging models, along with relevant clinical data. This multimodal approach will be trained and validated using a multi-institutional dataset.
Expected Outcomes: A robust, multimodal diagnostic model that outperforms single-modality models, offering a powerful tool for lung cancer diagnosis and prognosis.

Aim 4: Validate the multimodal AI diagnostic model using external test sets and clinical data.
Objective: To ensure the model's generalizability and clinical utility by testing it on independent datasets and real-world clinical data.
Methods: We will use external test sets from different institutions and integrate clinical data to validate the model's performance. Statistical analyses will be conducted to assess the model's accuracy, sensitivity, specificity, and predictive power.
Expected Outcomes: A clinically validated AI diagnostic model that can be implemented in healthcare settings to support lung cancer diagnosis and improve patient outcomes.

Collaborators

Currently only recruiting myself