Comparing and combining algorithms for computer-aided detection of pulmonary nodules in computed tomography scans: The ANODE09 study

Abstract

Numerous publications and commercial systems are available that deal with automatic detection of pulmonary nodules in thoracic computed tomography scans, but a comparative study where many systems are applied to the same data set has not yet been performed. This paper introduces ANODE09 ( http://anode09.isi.uu.nl), a database of 55 scans from a lung cancer screening program and a web-based framework for objective evaluation of nodule detection algorithms. Any team can upload results to facilitate benchmarking. The performance of six algorithms for which results are available are compared; five from academic groups and one commercially available system. A method to combine the output of multiple systems is proposed. Results show a substantial performance difference between algorithms, and demonstrate that combining the output of algorithms leads to marked performance improvements.

Automatic 3D pulmonary nodule detection in CT images: A survey

“…Generally, these databases are used to train students, to serve as a repository for rare cases, and to allow comparisons between the performance of different CADe systems [61]. Among the more important public databases available are: Lung Image Database Consortium (LIDC) [62,63], Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) [64,65], Early Lung Cancer Action Program (ELCAP) [7], Nederlands Leuvens Longkanker Screeningsonderzoek (NELSON) [66] and Automatic Nodule Detection 2009 (ANODE09) [67,68].…”

“…The ANODE09 is a database of 55 scans from a lung cancer screening program and a web-based framework for objective evaluation of nodule detection algorithms. Any team can upload results to facilitate benchmarking [66,64,68].…”

Sparse Classification for Computer Aided Diagnosis Using Learned Dictionaries

“…Our datasets are representative, but very challenging with large within-class variations for polyp, nodule class and other anatomical structures in colon and lung volumes. The results validate that this new classification framework can significantly improve the accuracy of our baseline computer-aided detection system, using the same set of input image features, and compare favorably with other state-of-the-arts [1][2][3][4][6][7][8].…”

“…However, these two cancers are highly preventable or "curable" if detected early. Image interpretation based cancer detection via 3D computer tomography has emerged as a common clinical practice, and many computer-aided detection tools for enhancing radiologists' diagnostic performance and effectiveness are developed in the last decade [1][2][3][4][6][7][8]. The key for radiologists to accept the clinical usage of a CAD system is highest possible detection sensitivity with reasonably low false positive (FP) rate per case.…”

“…more than 100 candidates per scan with one to two true positives. Then dozens or hundreds of heterogeneous image features can be computed per VOI, in domains of volumetric shape, intensity, gradient, texture and even context [1,3,4,7,8]. Lastly, the essential goal for classification is achieving the best balance between high sensitivities and low false positive rates, given VOIs and associated features.…”