Abstract:Medical imaging is vital in computer assisted intervention. Particularly cone beam computed tomography (CBCT) with defacto real time and mobility capabilities plays an important role. However, CBCT images often suffer from artifacts, which pose challenges for accurate interpretation, motivating research in advanced algorithms for more effective use in clinical practice. In this work we present CBCTLiTS, a synthetically generated, labelled CBCT dataset for segmentation with paired and aligned, high quality computed tomography data. The CBCT data is provided in 5 different levels of quality, reaching from a large number of projections with high visual quality and mild artifacts to a small number of projections with severe artifacts. This allows thorough investigations with the quality as a degree of freedom. We also provide baselines for several possible research scenarios like uni- and multimodal segmentation, multitask learning and style transfer followed by segmentation of relatively simple, liver to complex liver tumor segmentation. CBCTLiTS is accesssible via <a class="link-external link-https" href="https://www.kaggle.com/datasets/maximiliantschuchnig/cbct-liver-and-liver-tumor-segmentation-train-data" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to improve the application effect of cone - beam computed tomography (CBCT) images in medical image analysis, especially the performance in segmentation tasks. Specifically, the paper points out that due to its fast imaging and mobility, CBCT images play an important role in intraoperative imaging, but these images are often affected by artifacts, which pose a challenge to accurate image interpretation. To meet this challenge, the authors developed a synthetic, paired CBCT/CT dataset (CBCTLiTS), aiming to promote research on segmentation and style transfer. ### Main Objectives: 1. **Provide High - Quality Datasets**: Create a dataset containing CBCT images of different quality levels, each level ranging from a large number of projections with high visual quality and slight artifacts to a small number of projections with severe artifacts. This allows researchers to conduct thorough research under different image qualities. 2. **Support Multiple Research Scenarios**: Provide baseline results to support research scenarios such as unimodal and multimodal segmentation, multi - task learning, and style transfer. 3. **Promote Intraoperative Imaging Research**: Promote the development of intraoperative imaging technology by providing high - quality paired CBCT and CT data, especially in the application of liver and liver tumor segmentation. ### Specific Contributions: - **Dataset Creation**: Developed the CBCTLiTS dataset, including CBCT images of different quality levels, real - life segmentation masks, and paired CT samples. - **Baseline Results**: Provided baseline results for multiple research directions, including unimodal and multimodal segmentation, multi - task learning, and style transfer. ### Research Background: - **The Importance of CBCT in Intraoperative Imaging**: CBCT is very useful in intraoperative imaging because of its real - time and mobility, but the low image quality affects the performance of subsequent tasks (such as segmentation). - **The Deficiencies of Existing Datasets**: There are few publicly available CBCT datasets with real - life annotations, which limit the development and evaluation of automated CBCT image analysis methods. ### Experimental Setup: - **Segmentation Baseline**: Based on the 3D U - Net model, trained using binary cross - entropy and Dice similarity loss functions. - **Multi - task Learning**: Combine segmentation and image reconstruction tasks to explore whether it helps improve segmentation performance. - **Multimodal Learning**: Combine preoperative CT and intraoperative CBCT to explore the segmentation performance at different alignment levels. - **Style Transfer**: Convert CBCT images to the CT domain to remove structural artifacts and noise and improve segmentation performance. ### Results: - **CBCT Images of Different Quality Levels**: Show the changes in CBCT image quality under different numbers of projections. - **Segmentation Performance**: Provide the results of segmentation performance in different research scenarios, including unimodal and multimodal segmentation, multi - task learning, and style transfer. ### Discussion: - **The Uniqueness of the Dataset**: CBCTLiTS is a unique dataset that contains paired CBCT and CT data and is suitable for multiple research scenarios, especially research in the field of intraoperative imaging. - **The Advantages of Multimodal Learning**: Under the assumption of perfect alignment, multimodal learning can significantly improve segmentation performance, even exceeding the baseline CT model. - **The Potential of Style Transfer**: By converting CBCT images to the CT domain, artifacts and noise can be effectively removed and segmentation performance can be improved. ### Conclusion: The CBCTLiTS dataset provides a valuable resource for research on medical image segmentation and style transfer, especially for the development of intraoperative imaging technology. By providing CBCT images of different quality levels and paired CT data, this dataset supports multiple research scenarios and promotes further research in related fields.

CBCTLiTS: A Synthetic, Paired CBCT/CT Dataset For Segmentation And Style Transfer

Segmentation of Cardiac Tissues and Organs for CCTA Images Based on a Deep Learning Model

Initial Study On Improving Segmentation By Combining Preoperative CT And Intraoperative CBCT Using Synthetic Data

Multimodal Learning With Intraoperative CBCT & Variably Aligned Preoperative CT Data To Improve Segmentation

Multi-task Learning To Improve Semantic Segmentation Of CBCT Scans Using Image Reconstruction

A Benchmark Dataset for Segmenting Liver, Vasculature and Lesions from Large-scale Computed Tomography Data.

The KiTS19 Challenge Data: 300 Kidney Tumor Cases with Clinical Context, CT Semantic Segmentations, and Surgical Outcomes

Multi-scale anatomical regularization for domain-adaptive segmentation of pelvic CBCT images

Synthetic CT-aided Online CBCT Multi-Organ Segmentation for CBCT-guided Adaptive Radiotherapy of Pancreatic Cancer

MultiTalent: A Multi-Dataset Approach to Medical Image Segmentation

Multitask 3D CBCT‐to‐CT translation and organs‐at‐risk segmentation using physics‐based data augmentation

Synthetic CT generation from CBCT images via unsupervised deep learning

2DeteCT -- A large 2D expandable, trainable, experimental Computed Tomography dataset for machine learning

Validation and optimization of multi-organ segmentation on clinical imaging archives

Improving CT Image Segmentation Accuracy Using StyleGAN Driven Data Augmentation

Deep cross-modality (MR-CT) educed distillation learning for cone beam CT lung tumor segmentation

SegBook: A Simple Baseline and Cookbook for Volumetric Medical Image Segmentation

Automatic segmentation of kidney and liver tumors in CT images

TotalSegmentator: robust segmentation of 104 anatomical structures in CT images

SynthRAD2023 Grand Challenge dataset: generating synthetic CT for radiotherapy

ImageCAS: A Large-Scale Dataset and Benchmark for Coronary Artery Segmentation based on Computed Tomography Angiography Images