High-Fidelity Image Synthesis from Pulmonary Nodule Lesion Maps using Semantic Diffusion Model

Xuan Zhao,Benjamin Hou

2023-05-02

Abstract:Lung cancer has been one of the leading causes of cancer-related deaths worldwide for years. With the emergence of deep learning, computer-assisted diagnosis (CAD) models based on learning algorithms can accelerate the nodule screening process, providing valuable assistance to radiologists in their daily clinical workflows. However, developing such robust and accurate models often requires large-scale and diverse medical datasets with high-quality annotations. Generating synthetic data provides a pathway for augmenting datasets at a larger scale. Therefore, in this paper, we explore the use of Semantic Diffusion Mod- els (SDM) to generate high-fidelity pulmonary CT images from segmentation maps. We utilize annotation information from the LUNA16 dataset to create paired CT images and masks, and assess the quality of the generated images using the Frechet Inception Distance (FID), as well as on two common clinical downstream tasks: nodule detection and nodule localization. Achieving improvements of 3.96% for detection accuracy and 8.50% for AP50 in nodule localization task, respectively, demonstrates the feasibility of the approach.

Image and Video Processing,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper aims to solve the problem of insufficient data in lung nodule detection and localization. Specifically, the author explores the use of Semantic Diffusion Models (SDM) to generate high - fidelity lung CT images from lung nodule lesion maps in order to enhance the existing medical data sets. The synthetic data generated by this method can improve the performance of computer - aided diagnosis (CAD) models, especially in nodule detection and localization tasks. The paper mentions that the images generated using SDM perform well in downstream tasks, and compared with the baseline model and other methods (such as SPADE), there are significant improvements in detection accuracy and localization precision. The key points of the paper include: - **Problem background**: Lung cancer is one of the leading causes of cancer - related deaths worldwide. With the development of deep - learning technology, computer - aided diagnosis models based on learning algorithms can accelerate the nodule screening process and provide valuable assistance to radiologists. - **Data requirements**: Developing such robust and accurate models usually requires large - scale and diverse medical data sets with high - quality annotations. However, it is very difficult to obtain such data sets. - **Solution**: The author proposes to use SDM to generate high - fidelity lung CT images from segmentation maps to increase the size and diversity of the data set. By using the annotation information in the LUNA16 data set, paired CT images and masks are created, and the Fréchet Inception Distance (FID) is used to evaluate the quality of the generated images. - **Experimental results**: In two common clinical downstream tasks (nodule detection and nodule localization), after adding the images generated by SDM, the detection accuracy is increased by 3.96%, and the AP50 (average precision) is increased by 8.50% in the nodule localization task. This shows that this method is feasible in practical applications. Through these studies, the author demonstrates the potential of SDM in generating high - quality synthetic medical images, providing a new direction for future research and applications.

High-Fidelity Image Synthesis from Pulmonary Nodule Lesion Maps using Semantic Diffusion Model

Image Synthesis with Disentangled Attributes for Chest X-Ray Nodule Augmentation and Detection

A Synthesizing Semantic Characteristics Lung Nodules Classification Method Based on 3D Convolutional Neural Network.

Computer Aided Lung Cancer Diagnosis with Deep Learning Algorithms

Pulmonary Nodule Segmentation with CT Sample Synthesis Using Adversarial Networks

Robust and accurate pulmonary nodule detection with self-supervised feature learning on domain adaptation

A Survey of Computer-Aided Diagnosis of Lung Nodules from CT Scans Using Deep Learning

An Interpretable Three-Dimensional Artificial Intelligence Model for Computer-Aided Diagnosis of Lung Nodules in Computed Tomography Images

A Dual-Branch Framework With Prior Knowledge for Precise Segmentation of Lung Nodules in Challenging CT Scans

Realistic Lung Nodule Synthesis with Multi-Target Co-Guided Adversarial Mechanism

Automatic lung nodule segmentation and intra-nodular heterogeneity image generation

Class-Aware Adversarial Lung Nodule Synthesis in CT Images

AI in Lung Health: Benchmarking Detection and Diagnostic Models Across Multiple CT Scan Datasets

Automated Pulmonary Nodule Detection in CT Images Using Deep Convolutional Neural Networks.

Medical AI for Early Detection of Lung Cancer: A Survey

Lung Nodule Segmentation Combining Sobel Operator and Mask R-CNN

Faithful learning with sure data for lung nodule diagnosis

Automatic Pulmonary Lobe Segmentation Using Deep Learning

Explainable AI for Lung Nodule Detection and Classification in CT Images

Improved Lung Nodule Diagnosis Accuracy Using Lung CT Images with Uncertain Class

Deep Learning-Based Lung Medical Image Recognition