New Lung Lesions in Low-dose CT: a newly annotated longitudinal dataset derived from the National Lung Screening Trial

The National Lung Screening Trial (NLST) is an influential publicly available medical image dataset that has fueled a breadth of work in lesion detection. Strengths of this trial include its multi-institutional nature and the standardization of data collection. One limitation of the original dataset, however, is the lack of image labels such as lesion annotations.

We introduce an annotated derivative dataset of 152 lung lesions in 126 scans from 122 participants. These lesions were identified by radiologists during the trial as new compared to prior imaging and suspicious for malignancy. We provide point annotations, image coordinates, and registered baseline CT images for each identified new lesion. This addresses a gap in labeled longitudinal public datasets that capture the development of new lesions, and supports the development of automated tools in lesion detection that leverage temporal information in longitudinal imaging studies.