pith. sign in

arxiv: 1907.08328 · v1 · pith:XAVR5MIXnew · submitted 2019-07-19 · 📡 eess.IV · cs.CV

A multiscale Laplacian of Gaussian (LoG) filtering approach to pulmonary nodule detection from whole-lung CT scans

Sergei V. Fotin , David F. Yankelevitz , Claudia I. Henschke , Anthony P. Reeves This is my paper

Pith reviewed 2026-05-24 19:22 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords pulmonary nodule detectionLaplacian of GaussianCT scanscomputer-aided detectionmultiscale filteringsolid nodulesnonsolid nodulescandidate generation
0
0 comments X

The pith

A multiscale Laplacian of Gaussian filter detects 99.8 percent of solid pulmonary nodules in whole-lung CT scans.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a multiscale scale-normalized Laplacian of Gaussian filtering method to generate candidate locations for pulmonary nodules as the first stage of computer-aided detection systems. The approach scans entire low-dose CT volumes and supplies size estimates for each candidate. It was validated on a database of 706 scans containing 499 solid nodules at least 4 mm in diameter and 107 nonsolid nodules at least 6 mm in diameter. The method reached a sensitivity of 0.998 for the solid nodules and 1.000 for the nonsolid nodules, with average size errors of 0.12 mm and 1.27 mm respectively. A reader would care because reliable candidate generation can support faster and more complete initial searches in lung cancer screening workflows.

Core claim

The multiscale scale-normalized Laplacian of Gaussian filtering method creates a list of nodule candidate locations and sizes by detecting blob-like structures across scales. When applied to a size-enriched database of 706 whole-lung low-dose CT scans that contained 499 solid nodules of 4 mm or larger and 107 nonsolid nodules of 6 mm or larger, the method achieved sensitivities of 0.998 and 1.000. It also produced average size estimation errors of 0.12 mm for solid nodules and 1.27 mm for nonsolid nodules, together with average centroid localization distances of 1.41 mm and 1.43 mm compared with radiologist annotations.

What carries the argument

Multiscale scale-normalized Laplacian of Gaussian (LoG) filtering, which applies normalized second-derivative filters at multiple scales to locate and size blob-like intensity patterns that correspond to nodules.

If this is right

  • The LoG method supplies the initial candidate list and size estimates needed by later stages of a CAD pipeline.
  • The same filter set works for both solid and nonsolid nodule types at the reported size thresholds.
  • Size and centroid outputs are close enough to radiologist values to serve as starting points for further analysis.
  • High sensitivity at the candidate stage reduces the risk that a nodule is omitted from the final CAD output.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the reported sensitivities hold on unenriched consecutive screening cases, the method could be inserted directly into existing CAD workflows without retraining.
  • The scale-normalized LoG approach may transfer to detection of other approximately spherical lesions in volumetric CT or MR data.
  • Accurate automatic centroid placement could support longitudinal tracking of nodule growth across multiple scans.

Load-bearing premise

The size-enriched validation database represents the range of nodule sizes and appearances that appear in routine low-dose CT screening populations.

What would settle it

Running the same multiscale LoG method on a large consecutive series of unenriched screening CT scans and measuring whether sensitivity for solid nodules of 4 mm or larger falls below 0.99.

Figures

Figures reproduced from arXiv: 1907.08328 by Anthony P. Reeves, Claudia I. Henschke, David F. Yankelevitz, Sergei V. Fotin.

Figure 1
Figure 1. Figure 1: Appearance of a solid (a) and nonsolid (b) pulmonary n [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Rectangular function as one-dimensional represent [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: The maximum response occurs when the LoG kernel [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of multiscale LoG filtering: normalize [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of multiscale LoG filtering with respec [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Response of the normalized LoG kernels with differen [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: Selection of operating quantization step. Solid cur [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Exponentially increasing scale σi+1 = kσi results in the reduction in response bounded from below by Rdip. Rdip of the filter to a spherical model. Clearly, the minimum value will be reached when the responses from both scales are the same: Rdip = L(σ1, ddip) = L(σ2, ddip). (10) The nontrivial solution to this equation with respect to ddip is: ddip = √ 8σ1σ2 s ln σ2 3 − ln σ1 3 σ2 2 − σ1 2 , (11) which res… view at source ↗
Figure 9
Figure 9. Figure 9: Underestimation and overestimation of the sphere si [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Relative error in solid sphere size overestimation [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Central slice of the three-dimensional normalized [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Sphere-wall interference. Effect of the distance b [PITH_FULL_IMAGE:figures/full_fig_p008_14.png] view at source ↗
Figure 16
Figure 16. Figure 16: If the filter response at a considered point was maximal, i.e. it was greater or equal to the responses of all other neighboring points, the point was included to the set of nodule candidates. From the implementation convenience, instead of performing full four-dimensional search, the scales were processed in sequence while keeping in memory only the current scale and one scale above and below. The fourth… view at source ↗
Figure 15
Figure 15. Figure 15: Response functions at different scales. Shown is th [PITH_FULL_IMAGE:figures/full_fig_p010_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Four-dimensional neighborhood around a sample poi [PITH_FULL_IMAGE:figures/full_fig_p010_16.png] view at source ↗
Figure 18
Figure 18. Figure 18: Illustration of the interference effect between so [PITH_FULL_IMAGE:figures/full_fig_p011_18.png] view at source ↗
Figure 17
Figure 17. Figure 17: An output of the generator shown on one of the slices o [PITH_FULL_IMAGE:figures/full_fig_p011_17.png] view at source ↗
Figure 19
Figure 19. Figure 19: Image windowing technique. Intensity transform fu [PITH_FULL_IMAGE:figures/full_fig_p012_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Result of windowing on a local nodule subregion at a l [PITH_FULL_IMAGE:figures/full_fig_p012_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Distributions of solid (a) and nonsolid (b) nodule e [PITH_FULL_IMAGE:figures/full_fig_p013_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Distribution of the candidate normalized response [PITH_FULL_IMAGE:figures/full_fig_p014_22.png] view at source ↗
Figure 7
Figure 7. Figure 7: The graph has three local modes that were visually [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 23
Figure 23. Figure 23: Distribution of candidate sizes before and after su [PITH_FULL_IMAGE:figures/full_fig_p014_23.png] view at source ↗
read the original abstract

Candidate generation, the first stage for most computer aided detection (CAD) systems, rapidly scans the entire image data for any possible abnormality locations, while the subsequent stages of the CAD system refine the candidates list to determine the most probable or significant of these candidates. The candidate generator creates a list of the locations and provides a size estimate for each candidate. A multiscale scale-normalized Laplacian of Gaussian (LoG) filtering method for detecting pulmonary nodules in whole-lung CT scans, presented in this paper, achieves a high sensitivity for both solid and nonsolid pulmonary nodules. The pulmonary nodule LoG filtering method was validated on a size-enriched database of 706 whole-lung low-dose CT scans containing 499 solid (>= 4 mm) and 107 nonsolid (>= 6 mm) pulmonary nodules. The method achieved a sensitivity of 0.998 (498/499) for solid nodules and a sensitivity of 1.000 (107/107) for nonsolid nodules. Furthermore, compared to radiologist measurements, the method provided low average nodule size estimation error of 0.12 mm for solid and 1.27 mm for nonsolid nodules. The average distance between automatically and manually determined nodule centroids were 1.41 mm and 1.43 mm, respectively.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents a multiscale scale-normalized Laplacian of Gaussian (LoG) filtering method for candidate generation in pulmonary nodule detection from whole-lung low-dose CT scans. Validation is reported on a size-enriched database of 706 scans containing 499 solid (>=4 mm) and 107 nonsolid (>=6 mm) nodules, achieving sensitivities of 0.998 (498/499) and 1.000 (107/107), respectively, with average size estimation errors of 0.12 mm (solid) and 1.27 mm (nonsolid) and centroid localization errors of 1.41 mm and 1.43 mm.

Significance. If the performance generalizes, the method supplies a straightforward scale-space technique for high-sensitivity candidate generation that handles both solid and nonsolid nodules and supplies usable size and position estimates. The sizable nodule count in the validation set is a strength for a candidate-generation paper.

major comments (2)
  1. [Abstract] Abstract: The validation set is labeled 'size-enriched' with no description of the enrichment procedure, no nodule size or attenuation histograms, and no comparison to an unenriched screening cohort. This detail is load-bearing for the central sensitivity claims (0.998/1.000), because enrichment may preferentially retain nodules whose scale and contrast align with the chosen LoG kernels, limiting claims of general performance on routine screening populations.
  2. [Abstract] Abstract: No false-positive rate, candidates per scan, or operating-point trade-off is supplied. For a candidate-generation stage the sensitivity-specificity balance is essential to judge downstream CAD utility; its absence prevents assessment of whether the reported sensitivities are achieved at a usable false-positive burden.
minor comments (1)
  1. [Abstract] Abstract: The larger size-estimation error for nonsolid nodules (1.27 mm) versus solid (0.12 mm) is reported without comment; a short explanation of the difference would aid interpretation.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the thorough review and valuable comments on our manuscript. We address each of the major comments in detail below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The validation set is labeled 'size-enriched' with no description of the enrichment procedure, no nodule size or attenuation histograms, and no comparison to an unenriched screening cohort. This detail is load-bearing for the central sensitivity claims (0.998/1.000), because enrichment may preferentially retain nodules whose scale and contrast align with the chosen LoG kernels, limiting claims of general performance on routine screening populations.

    Authors: The term 'size-enriched' indicates that the database was constructed to include nodules spanning a range of sizes to allow evaluation across different scales. We agree that more details would be beneficial. In the revised manuscript, we will expand the description of the database construction to include the enrichment procedure and provide histograms of nodule sizes and attenuations. A comparison to an unenriched screening cohort is not available from the current dataset. revision: partial

  2. Referee: [Abstract] Abstract: No false-positive rate, candidates per scan, or operating-point trade-off is supplied. For a candidate-generation stage the sensitivity-specificity balance is essential to judge downstream CAD utility; its absence prevents assessment of whether the reported sensitivities are achieved at a usable false-positive burden.

    Authors: We acknowledge the importance of reporting the false-positive characteristics for a candidate generation method. The manuscript prioritizes demonstrating the high sensitivity and accurate size/centroid estimation. We will revise the manuscript to include the average number of candidates per scan and discuss the implications for downstream processing. revision: yes

standing simulated objections not resolved
  • Comparison of the method's performance on an unenriched routine screening population is not possible with the available data.

Circularity Check

0 steps flagged

No significant circularity; validation metrics are direct counts against external annotations

full rationale

The paper describes a standard multiscale LoG filter for candidate generation and reports sensitivity as the direct ratio of detected nodules to the total number of manually annotated nodules in the 706-scan database. No equations, parameters, or predictions are shown to be fitted from the evaluation set and then re-used to compute the same sensitivities. The size-enrichment issue raised by the skeptic affects external validity but does not create a self-referential derivation inside the paper. No self-citation chains or ansatzes are invoked to justify the core detection step.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that pulmonary nodules produce blob-like intensity profiles detectable by LoG filters and that the chosen validation set adequately represents clinical variability; no new entities are postulated and no free parameters are explicitly fitted in the abstract.

axioms (1)
  • domain assumption Pulmonary nodules appear as approximately Gaussian blobs in CT intensity volumes at clinically relevant scales.
    Implicit justification for choosing the LoG filter family.

pith-pipeline@v0.9.0 · 5784 in / 1299 out tokens · 24631 ms · 2026-05-24T19:22:10.750041+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Detection of pulmonary nodules at multirow-detector CT: e ffectiveness of double reading to improve sensitivity at standard-dose a nd low-dose chest CT,

    D. Wormanns, K. Ludwig, F. Beyer, W. Heindel, and S. Diede rich, “Detection of pulmonary nodules at multirow-detector CT: e ffectiveness of double reading to improve sensitivity at standard-dose a nd low-dose chest CT,” European radiology, vol. 15, no. 1, pp. 14–22, 2005

    work page 2005

  2. [2]

    Pulmonary nodule det ection with low-dose CT of the lung: agreement among radiologists,

    J. K. Leader, T. E. Warfel, C. R. Fuhrman, S. K. Golla, J. L. Weissfeld, R. S. Avila, W. D. Turner, and B. Zheng, “Pulmonary nodule det ection with low-dose CT of the lung: agreement among radiologists, ” American Journal of Roentgenology , vol. 185, no. 4, p. 973, 2005

    work page 2005

  3. [3]

    Computer-aided detection of pulmonary nodules: influence of nodule characteristics on detection performance,

    K. Marten, C. Engelke, T. Seyfarth, A. Grillhosl, S. Oben auer, and E. J. Rummeny, “Computer-aided detection of pulmonary nodules: influence of nodule characteristics on detection performance,” Clinical Radiology, vol. 60, no. 2, pp. 196–206, 2005

    work page 2005

  4. [4]

    Sm all pulmonary nodules: effect of two computer-aided detection systems on radiologist performance,

    M. Das, G. Muhlenbruch, A. H. Mahnken, T. G. Flohr, L. Gund el, S. Stanzel, T. Kraus, R. W. Gunther, and J. E. Wildberger, “Sm all pulmonary nodules: effect of two computer-aided detection systems on radiologist performance,” Radiology, vol. 241, no. 2, p. 564, 2006

    work page 2006

  5. [5]

    Comparison of sensitivity and reading time for the use of computer-aided detection (CAD) of pulmon ary nodules at MDCT as concurrent or second reader,

    F. Beyer, L. Zierott, E. M. Fallenberg, K. U. Juergens, J. Stoeckel, W. Heindel, and D. Wormanns, “Comparison of sensitivity and reading time for the use of computer-aided detection (CAD) of pulmon ary nodules at MDCT as concurrent or second reader,” European Radiology, vol. 17, no. 11, pp. 2941–2947, 2007

    work page 2007

  6. [6]

    Computer analysis of computed tomography scans of the lung: a survey,

    I. Sluimer, A. Schilham, M. Prokop, and B. van Ginneken, “ Computer analysis of computed tomography scans of the lung: a survey, ” IEEE Transactions on Medical Imaging , vol. 25, no. 4, pp. 385–405, April 2006

    work page 2006

  7. [7]

    Recent progress in computer-aided diagnosis of l ung nodules on thin-section CT,

    Q. Li, “Recent progress in computer-aided diagnosis of l ung nodules on thin-section CT,” Computerized Medical Imaging and Graphics , vol. 31, no. 4-5, pp. 248–257, 2007. 16

    work page 2007

  8. [8]

    Comput er-Aided Diagnosis of Lung Cancer and Pulmonary Embolism in Computed Tomography – A Review,

    H. P . Chan, L. Hadjiiski, C. Zhou, and B. Sahiner, “Comput er-Aided Diagnosis of Lung Cancer and Pulmonary Embolism in Computed Tomography – A Review,” Academic Radiology, vol. 15, no. 5, pp. 535– 555, 2008

    work page 2008

  9. [9]

    Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm,

    B. Zhao, “Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm,” Journal of applied clinical medical physics, vol. 4, no. 3, p. 248, 2003

    work page 2003

  10. [10]

    Automatic detection and recognition of lung abnormalitie s in helical CT images using deformable templates,

    A. A. Farag, A. El-Baz, G. G. Gimel farb, R. Falk, and S. G. Hushek, “Automatic detection and recognition of lung abnormalitie s in helical CT images using deformable templates,” Lecture Notes in Computer Science, pp. 856–864, 2004

    work page 2004

  11. [11]

    Evaluation of automated lung nodule detec tion on low-dose computed tomography scans from a lung cancer scr eening program,

    S. G. Armato, A. S. Roy, H. Macmahon, F. Li, K. Doi, S. Sone , and M. B. Altman, “Evaluation of automated lung nodule detec tion on low-dose computed tomography scans from a lung cancer scr eening program,” Academic Radiology, vol. 12, no. 3, pp. 337–46, March 2005

    work page 2005

  12. [12]

    Automated Detection of Small Solid Pulmonary Nodules in Wh ole Lung CT Scans from a Lung Cancer Screening Study,

    A. Enquobahrie, A. P . Reeves, D. F. Y ankelevitz, and C. I . Henschke, “Automated Detection of Small Solid Pulmonary Nodules in Wh ole Lung CT Scans from a Lung Cancer Screening Study,” Academic Radiology, vol. 14, no. 5, pp. 579–593, 2007

    work page 2007

  13. [13]

    An au tomated CT based lung nodule detection scheme using geometric analy sis of signed distance field,

    J. Pu, B. Zheng, J. K. Leader, X. H. Wang, and D. Gur, “An au tomated CT based lung nodule detection scheme using geometric analy sis of signed distance field,” Medical Physics , vol. 35, p. 3453, 2008

    work page 2008

  14. [14]

    Computerized Detection of Lung Nodules in Thin-Section CT Images by Use of Selective Enhancement Fi lters and an Automated Rule-Based Classifier,

    Q. Li, F. Li, and K. Doi, “Computerized Detection of Lung Nodules in Thin-Section CT Images by Use of Selective Enhancement Fi lters and an Automated Rule-Based Classifier,” Academic radiology, vol. 15, no. 2, pp. 165–175, 2008

    work page 2008

  15. [15]

    A large-scale evaluation of automati c pulmonary nodule detection in chest CT using local image features and k -nearest- neighbour classification,

    K. Murphy, B. van Ginneken, A. M. Schilham, B. J. de Hoop, H. A. Gi- etema, and M. Prokop, “A large-scale evaluation of automati c pulmonary nodule detection in chest CT using local image features and k -nearest- neighbour classification,” Medical image analysis , vol. 13, no. 5, p. 757, 2009

    work page 2009

  16. [16]

    A new computationa lly efficient cad system for pulmonary nodule detection in ct imagery,

    T. Messay, R. Hardie, and S. Rogers, “A new computationa lly efficient cad system for pulmonary nodule detection in ct imagery,” Medical image analysis , vol. 14, no. 3, pp. 390–406, 2010

    work page 2010

  17. [17]

    Theory of edge detection,

    D. Marr and E. Hildreth, “Theory of edge detection,” Proceedings of the Royal Society of London. Series B, Biological Sciences , pp. 187–217, 1980

    work page 1980

  18. [18]

    Identification and boundar y extraction of blobs in complex imagery,

    T. Jiang and M. B. Merickel, “Identification and boundar y extraction of blobs in complex imagery,” Computerized Medical Imaging and Graphics, vol. 13, no. 5, pp. 369–382, 1989

    work page 1989

  19. [19]

    Shape from texture: Integrat ing texture- element extraction and surface estimation,

    D. Blostein and N. Ahuja, “Shape from texture: Integrat ing texture- element extraction and surface estimation,” Pattern Analysis and Ma- chine Intelligence, IEEE Transactions on , vol. 11, no. 12, pp. 1233– 1251, 1989

    work page 1989

  20. [20]

    Detecting salient blob-like image stru ctures and their scales with a scale-space primal sketch: A method for focus- of- attention,

    T. Lindeberg, “Detecting salient blob-like image stru ctures and their scales with a scale-space primal sketch: A method for focus- of- attention,” International Journal of Computer Vision , vol. 11, no. 4, pp. 283–318, 1993

    work page 1993

  21. [21]

    Multi-s cale Nodule Detection in Chest Radiographs,

    A. M. R. Schilham, B. van Ginneken, and M. Loog, “Multi-s cale Nodule Detection in Chest Radiographs,” LECTURE NOTES IN COMPUTER SCIENCE, pp. 602–609, 2003

    work page 2003

  22. [22]

    On measuring the change in size of pulmonar y nodules,

    A. P . Reeves, A. B. Chan, D. F. Y ankelevitz, C. I. Henschk e, B. Kressler, and W. J. Kostis, “On measuring the change in size of pulmonar y nodules,” Medical Imaging, IEEE Transactions on , vol. 25, no. 4, pp. 435–450, 2006

    work page 2006

  23. [23]

    Automated nodule locatio n and size estimation using a multi-scale laplacian of gaussian filter ing approach,

    A. C. Jirapatnakul, S. V . Fotin, A. P . Reeves, A. M. Bianc ardi, D. F. Y ankelevitz, and C. I. Henschke, “Automated nodule locatio n and size estimation using a multi-scale laplacian of gaussian filter ing approach,” in 31st Annual International Conference of the IEEE Engineeri ng in Medicine and Biology Society , Sept. 2009, pp. 1028–1031

    work page 2009

  24. [24]

    The LoG Characteristic Scale: A Consistent M easure- ment of Lung Nodule Size in CT Imaging

    S. Diciotti, S. Lombardo, G. Coppini, L. Grassi, M. Falc hini, and M. Mascalchi, “The LoG Characteristic Scale: A Consistent M easure- ment of Lung Nodule Size in CT Imaging.” IEEE transactions on medical imaging , vol. 29, no. 2, p. 397, 2010

    work page 2010

  25. [25]

    Automatic Two-Step Detection of Pulmonary Nodules,

    M. Dolejsi and J. Kybic, “Automatic Two-Step Detection of Pulmonary Nodules,” in Proceedings of SPIE , vol. 6514, 2007, p. 65143J

    work page 2007

  26. [26]

    The segmentation of nonsolid pulmonary no dules in CT images,

    W. Browder, “The segmentation of nonsolid pulmonary no dules in CT images,” Master’s thesis, Cornell University, 2007

    work page 2007

  27. [27]

    Glo ssary of terms for CT of the lungs: recommendations of the Nomencla ture Committee of the Fleischner Society,

    J. H. M. Austin, N. L. Muller, P . J. Friedman, D. M. Hansel l, D. P . Naidich, M. Remy-Jardin, W. R. Webb, and E. A. Zerhouni, “Glo ssary of terms for CT of the lungs: recommendations of the Nomencla ture Committee of the Fleischner Society,” Radiology, vol. 200, no. 2, pp. 327–332, 1996

    work page 1996

  28. [28]

    Early Lung Cancer Action Project: overall design and findings from baseline screening,

    C. I. Henschke, D. I. McCauley, D. F. Y ankelevitz, D. P . N aidich, G. McGuinness, O. S. Miettinen, D. M. Libby, M. W. Pasmantier , J. Koizumi, N. K. Altorki et al. , “Early Lung Cancer Action Project: overall design and findings from baseline screening,” The Lancet , vol. 354, no. 9173, pp. 99–105, 1999

    work page 1999

  29. [29]

    Nodule management protocol of the NELSON randomised lung c ancer screening trial,

    D. M. Xu, H. Gietema, H. de Koning, R. V ernhout, K. Nackae rts, M. Prokop, C. Weenink, J. W. Lammers, H. Groen, M. Oudkerk et al. , “Nodule management protocol of the NELSON randomised lung c ancer screening trial,” Lung Cancer , vol. 54, no. 2, pp. 177–184, 2006

    work page 2006

  30. [30]

    Design, recruitment and baseline results of the ITALUNG tr ial for lung cancer screening with low-dose CT,

    A. Lopes Pegna, G. Picozzi, M. Mascalchi, F. Maria Caroz zi, L. Car- rozzi, C. Comin, C. Spinelli, F. Falaschi, M. Grazzini, F. In nocenti et al., “Design, recruitment and baseline results of the ITALUNG tr ial for lung cancer screening with low-dose CT,” Lung Cancer , vol. 64, no. 1, pp. 34–40, 2009

    work page 2009

  31. [31]

    Early processing of visual information,

    D. Marr, “Early processing of visual information,” Philosophical Trans- actions of the Royal Society of London. Series B, Biological Sciences, vol. 275, no. 942, pp. 483–519, 1976

    work page 1976

  32. [32]

    Feature detection with automatic scale selection,

    T. Lindeberg, “Feature detection with automatic scale selection,” Inter- national Journal of Computer Vision , vol. 30, no. 2, pp. 79–116, 1998

    work page 1998

  33. [33]

    Automated volumetric seg mentation method for growth consistency of nonsolid pulmonary nodule s in high- resolution CT,

    W. A. Browder, A. P . Reeves, T. V . Apananosovich, M. D. Ch am, D. F. Y ankelevitz, and C. I. Henschke, “Automated volumetric seg mentation method for growth consistency of nonsolid pulmonary nodule s in high- resolution CT,” in Proc. of SPIE V ol , vol. 6514, 2007, pp. 65 140Y –1

    work page 2007

  34. [34]

    The design and implementati on of FFTW3,

    M. Frigo and S. G. Johnson, “The design and implementati on of FFTW3,” Proceedings of the IEEE , vol. 93, no. 2, pp. 216–231, 2005

    work page 2005

  35. [35]

    Recursive implementation of LoG fil tering,

    J. S. Jin and Y . Gao, “Recursive implementation of LoG fil tering,” Real- Time Imaging, vol. 3, no. 1, pp. 59–65, 1997

    work page 1997