ABSTRACT
RATIONALE AND OBJECTIVES: To analyze radiologist lung nodule segmentations in the Lung Imaging Database Consortium (LIDC) database and to apply statistical tools to generate estimates of ground truth. This investigation expands on earlier work by considering a larger number of cases from the LIDC database, and results were generated on a per-nodule basis, as opposed to a per-case basis as was done previously. MATERIALS AND METHODS: We analyzed nodule data drawn from the 41 most recent computed tomography exams released by the LIDC. We combined radiologist segmentations for a given nodule using different consensus schemes: union, intersection, and simultaneous truth and performance level estimation (STAPLE). We also generated three-dimensional models of the manual segmentations using discrete marching cubes to visualize features of the data. RESULTS: Using the union as the consensus scheme produced the greatest number of nodule-positive voxels while using the intersection produced the fewest. Considering only nodules for which all readers agreed on nodule presence, STAPLE computed sensitivity averages for readers one, two, three, and four were 0.91, 0.83, 0.90, and 0.77, respectively. Specificity averages were 0.97, 0.98, 0.97, and 0.97. Considering cases for which there was disagreement about nodule presence, sensitivity results become 0.67, 0.74, 0.60, and 0.37. Specificity results in this case are 0.95, 0.95, 0.95, and 0.98. STAPLE-generated pmaps exhibited probability values tightly grouped below the 0.25 and above the 0.75 probability levels. Three-dimensional models of manually segmented nodules revealed step-artefacts in the segmentation data. CONCLUSIONS: Radiologists often disagree about nodule presence. Ideally, knowing each reader's sensitivity and specificity a priori is preferred for optimal STAPLE results. Knowing these values and developing manual segmentation tools and imaging protocols that mitigate unwanted segmentation features (such as step artefacts) can result in more accurate estimates of ground truth. Furthermore, a computer-aided detection algorithm's performance is a function of the ground truth estimate by which it is scored.
Subject(s)
Algorithms , Artificial Intelligence , Databases, Factual , Pattern Recognition, Automated/methods , Radiographic Image Interpretation, Computer-Assisted/methods , Solitary Pulmonary Nodule/diagnostic imaging , Humans , Lung Neoplasms/diagnostic imaging , Observer Variation , Radiographic Image Enhancement/methods , Reproducibility of Results , Sensitivity and Specificity , United StatesABSTRACT
Lung cancer remains an ongoing problem resulting in substantial deaths in the United States and the world. Within the United states, cancer of the lung and bronchus are the leading causes of fatal malignancy and make up 32% of the cancer deaths among men and 25% of the cancer deaths among women. Five year survival is low, (14%), but recent studies are beginning to provide some hope that we can increase survivability of lung cancer provided that the cancer is caught and treated in early stages. These results motivate revisiting the concept of lung cancer screening using thin slice multidetector computed tomography (MDCT) protocols and automated detection algorithms to facilitate early detection. In this environment, resources to aid Computer Aided Detection (CAD) researchers to rapidly develop and harden detection and diagnostic algorithms may have a significant impact on world health. The National Cancer Institute (NCI) formed the Lung Imaging Database Consortium (LIDC) to establish a resource for detecting, sizing, and characterizing lung nodules. This resource consists of multiple CT chest exams containing lung nodules that seveal radiologists manually countoured and characterized. Consensus on the location of the nodule boundaries, or even on the existence of a nodule at a particular location in the lung was not enforced, and each contour is considered a possible nodule. The researcher is encouraged to develop measures of ground truth to reconcile the multiple radiologist marks. This paper analyzes these marks to determine radiologist agreement and to apply statistical tools to the generation of a nodule ground truth. Features of the resulting consensus and individual markings are analyzed.