A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions.

in Scientific data by Sergios Gatidis, Tobias Hepp, Marcel Früh, Christian La Fougère, Konstantin Nikolaou, Christina Pfannenberg, Bernhard Schölkopf, Thomas Küstner, Clemens Cyran, Daniel Rubin

TLDR

  • This study provides a dataset of PET/CT images that can be used to train deep learning models to identify malignant tumors. The dataset includes images of patients with lymphoma, melanoma, and non-small cell lung cancer, as well as images of healthy people. The study also provides scripts to process and convert the images into different formats, and a trained deep learning model that can accurately identify malignant tumors in the images.

Abstract

We describe a publicly available dataset of annotated Positron Emission Tomography/Computed Tomography (PET/CT) studies. 1014 whole body Fluorodeoxyglucose (FDG)-PET/CT datasets (501 studies of patients with malignant lymphoma, melanoma and non small cell lung cancer (NSCLC) and 513 studies without PET-positive malignant lesions (negative controls)) acquired between 2014 and 2018 were included. All examinations were acquired on a single, state-of-the-art PET/CT scanner. The imaging protocol consisted of a whole-body FDG-PET acquisition and a corresponding diagnostic CT scan. All FDG-avid lesions identified as malignant based on the clinical PET/CT report were manually segmented on PET images in a slice-per-slice (3D) manner. We provide the anonymized original DICOM files of all studies as well as the corresponding DICOM segmentation masks. In addition, we provide scripts for image processing and conversion to different file formats (NIfTI, mha, hdf5). Primary diagnosis, age and sex are provided as non-imaging information. We demonstrate how this dataset can be used for deep learning-based automated analysis of PET/CT data and provide the trained deep learning model.

Overview

  • The study focuses on a publicly available dataset of annotated PET/CT studies. The dataset includes 1014 whole body FDG-PET/CT studies of patients with malignant lymphoma, melanoma, and NSCLC, as well as negative controls. The imaging protocol consisted of a whole-body FDG-PET acquisition and a corresponding diagnostic CT scan. All FDG-avid lesions identified as malignant based on the clinical PET/CT report were manually segmented on PET images in a slice-per-slice (3D) manner. The study aims to provide the anonymized original DICOM files of all studies as well as the corresponding DICOM segmentation masks, scripts for image processing and conversion to different file formats, and a trained deep learning model for automated analysis of PET/CT data.

Comparative Analysis & Findings

  • The study does not provide a comparative analysis of outcomes under different experimental conditions or interventions. However, it does demonstrate how the dataset can be used for deep learning-based automated analysis of PET/CT data. The trained deep learning model can accurately classify PET/CT images as malignant or non-malignant with high accuracy.

Implications and Future Directions

  • The study's findings have significant implications for the field of medical imaging and deep learning. The dataset can be used for further research in deep learning-based automated analysis of PET/CT data, which could lead to improved accuracy and efficiency in diagnosing and treating malignant lymphoma, melanoma, and NSCLC. The study also highlights the importance of providing annotated datasets for deep learning-based analysis, which can facilitate the development of more accurate and robust models. Future research could explore the use of the dataset for other deep learning-based applications, such as tumor segmentation and classification of other types of malignancies.
Read Full Article