Descriptor: Manually Annotated CT Dataset of Lung Lobes in COVID-19 and Cancer Patients (LOCCA)

The task of semantic segmentation of medical images involves classifying each pixel of an image into specific categories or regions of interest. The success of algorithms performing this task depends on the availability of high-quality image data, accompanied by corresponding annotations provided by experts. However, the number of publicly available annotations for pulmonary lobes in the literature is scarce, which hinders the development of automated segmentation methods. Furthermore, no public dataset was found containing CT images with manual annotation of the pulmonary lobes for patients presenting severe abnormalities. In this work, a novel collection of volumetric lung images and their corresponding annotations has been made available under an open-source license. The objective of the provided data collection is to encourage the development of novel lobe segmentation methods and allow other models to be evaluated specifically on images of severe pulmonary lesions. A collaborative effort between students and experts resulted in a dataset containing 30 computed tomography (CT) volumes of Brazilian COVID-19 patients presenting opacities and consolidations, accompanied by the corresponding manual annotations of each lung lobe. Additionally, lung lobe manual annotations were created for 30 publicly available CT volumes of cancer patients, sourced from the Medical Segmentation Decathlon, for a total of 60 volumetric manual lung lobe annotations. The data is provided together with characterization of abnormal finding severity for each image and lobe.