pith. sign in

arxiv: 1906.08620 · v2 · pith:Z4UBGAZGnew · submitted 2019-06-20 · 💻 cs.CV

BGrowth: an efficient approach for the segmentation of vertebral compression fractures in magnetic resonance imaging

Jonathan S. Ramos , Carolina Y. V. Watanabe , Marcello H. Nogueira-Barbosa , Agma J. M. Traina This is my paper

Pith reviewed 2026-05-25 19:37 UTC · model grok-4.3

classification 💻 cs.CV
keywords BGrowthvertebral compression fracturesMRI segmentationsemi-automatic segmentationmedical image analysisfracture segmentationimage segmentation
0
0 comments X

The pith

BGrowth segments vertebral compression fractures in MRI with up to 95% accuracy from rough seed points using balanced region growth.

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

The paper introduces BGrowth, a semi-automatic segmentation technique designed for MRI scans of crushed vertebral bodies. These fractures create regions of uneven intensity that resemble nearby structures, defeating many standard segmentation approaches. The method expands from user-provided seed points through a balanced growth process that accounts for intensity variation. On a set of 191 vertebrae it reaches 95% accuracy while matching the speed of existing methods and tolerating imprecise initial annotations. Accurate vertebral outlines matter because they feed into diagnosis, classification, and treatment planning for back pain.

Core claim

BGrowth is a semi-automatic segmentation method that significantly outperforms well-known methods from the literature, reaching an accuracy of up to 95% on a dataset with 102 crushed and 89 normal vertebrae. It maintains acceptable processing times and performs best even with rough manual annotation of seed points, addressing the challenges of non-homogeneous intensities in fractured regions.

What carries the argument

The balanced growth mechanism, which expands segmented regions while maintaining balance to handle intensity variations and dark regions similar to nearby structures.

If this is right

  • More accurate vertebral outlines improve downstream classification and quantitative analysis of fractures.
  • Tolerance for rough seed points reduces the time and expertise needed for manual initialization.
  • Processing time comparable to existing methods supports routine clinical use.
  • The approach works on both crushed and normal vertebrae within the same imaging protocol.

Where Pith is reading between the lines

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

  • Growth strategies that balance intensity statistics could transfer to segmentation tasks in other organs that exhibit similar inhomogeneity.
  • Pairing BGrowth with automated seed detection would move the pipeline closer to full automation.
  • Multi-center validation on varied scanner models would test whether the reported accuracy generalizes.

Load-bearing premise

The performance gains from balanced growth observed on this collection of 191 vertebrae will hold for other MRI scanners, patient populations, and fracture types.

What would settle it

Running BGrowth on an independent MRI dataset collected at a different site and finding its accuracy no higher than standard methods would falsify the superiority claim.

Figures

Figures reproduced from arXiv: 1906.08620 by Agma J. M. Traina, Carolina Y. V. Watanabe, Jonathan S. Ramos, Marcello H. Nogueira-Barbosa.

Figure 1
Figure 1. Figure 1: Example of manual vertebral bodies segmentation. els, region growing and Otsu threshold. Region growing techniques are used in (9), such as, snakes (Chan-Vese), Otsu and fuzzy c-means clustering, in order to compose a cooperative strategy for a dynamic ensemble of clas￾sification models. Although both works have presented very promising results, the VBSeg method presents a low segmentation performance on V… view at source ↗
Figure 2
Figure 2. Figure 2: Balanced Growth (BGrowth) method. Cut (16) (presents slightly lower segmentation perfor￾mance than the original GrowCut (16)), employ several seeds points inside and outside the object of interest and have been widely used for many medical MRI exams (especially in oncology) (17). However, to the best of our knowledge, GrowCut was not tested on VCFs. Based on the formulation of segmentation as an en￾ergy mi… view at source ↗
Figure 3
Figure 3. Figure 3: BGrowth’ iterations: Ground-Truth (GT), interior annotation, exterior annotation and final result boundaries are outlined in red, white, green and cyan, respectively. goes back and forth on the same pixel. Therefore, we would recommend the definition of a maximum num￾ber of iterations depending on the kind of image being segmented. We have empirically used a maximum of 30 iterations for the segmentation of… view at source ↗
Figure 4
Figure 4. Figure 4: BGrowth’ iterations on five lumbar: Ground-Truth, interior annotation and exterior annotation boundaries are out￾lined in red, white and green, respectively. 3.2 Comparison measures We analyzed the Jaccard Coefficient J and Dice Score D (24, 25): J(GT, Seg) = |GT ∩ Seg| |GT ∪ Seg| ; (3) D(GT, Seg) = 2 × |GT ∩ Seg| |GT| + |Seg| , (4) in which GT represents the ground-truth region and Seg represents the regi… view at source ↗
Figure 5
Figure 5. Figure 5: Example of sloppy annotations (seed points): ground-truth (GT) in red; interior and exterior annotations in white and green, respectively. impact of changing the percentage of manual annota￾tion inside and outside each vertebral body for every ap￾proach that uses this kind of annotation. 4.1 Overall measures analysis [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Segmentation results for a case with all lumbar vertebrae with malignant compression fractures. The percentages shown represents the Jaccard Coefficient. Ground-truth in red and semi-automatic segmentation in cyan. ‘All’, BGrowth presented significantly better results than LazySnapping for Jaccard, Dice and F-measure. BGrowth improved GrowCut’s performance in the three first cases (‘All’, ‘Normal’ and ‘Ben… view at source ↗
Figure 7
Figure 7. Figure 7: shows a few examples of variations of the interior percentages and the distances from the ground￾truth used [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of Jaccard coefficient over annotation variations at the interior and exterior of each vertebral body. ever, the F-measure is higher for BG, which implicates that, in general, BG presented a better balance between precision and recall than GC. Although LazySnapping (LS) presented higher precision most of the time, its re￾call is one of the lowest. On the other hand, GrabCut (GB) presented one of… view at source ↗
read the original abstract

Segmentation of medical images is a critical issue: several process of analysis and classification rely on this segmentation. With the growing number of people presenting back pain and problems related to it, the automatic or semi-automatic segmentation of fractured vertebral bodies became a challenging task. In general, those fractures present several regions with non-homogeneous intensities and the dark regions are quite similar to the structures nearby. Aimed at overriding this challenge, in this paper we present a semi-automatic segmentation method, called Balanced Growth (BGrowth). The experimental results on a dataset with 102 crushed and 89 normal vertebrae show that our approach significantly outperforms well-known methods from the literature. We have achieved an accuracy up to 95% while keeping acceptable processing time performance, that is equivalent to the state-of-the-artmethods. Moreover, BGrowth presents the best results even with a rough (sloppy) manual annotation (seed points).

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

3 major / 3 minor

Summary. The paper presents BGrowth, a semi-automatic region-growing segmentation method for vertebral compression fractures in MRI. It claims that the balanced-growth rule enables robust handling of intensity inhomogeneities, yielding up to 95% accuracy on a private dataset of 102 crushed and 89 normal vertebrae while outperforming several literature methods even when initialized with rough manual seeds and maintaining competitive run times.

Significance. A reliable, efficient semi-automatic tool for VCF segmentation would be useful for quantitative analysis of back-pain cohorts. The paper supplies no machine-checked proofs, open code, or parameter-free derivations, so any significance is conditional on the empirical claims; the current evaluation does not yet allow that assessment.

major comments (3)
  1. [§4 and §5] §4 (Experimental Setup) and §5 (Results): the central claim that BGrowth 'significantly outperforms' the compared methods rests on aggregate accuracy figures alone. No definition is given for the accuracy measure (Dice, Jaccard, pixel-wise, or Hausdorff), no per-method means ± std are reported, and no paired statistical test (Wilcoxon signed-rank or t-test) across the 191 cases is performed. This directly undermines the 'significantly outperforms' assertion.
  2. [§5] §5, Table 2 (or equivalent comparison table): the manuscript states that BGrowth remains best even with 'rough (sloppy) manual annotation,' yet provides no quantitative characterization of seed-point variability (number of seeds, placement tolerance, or inter-observer agreement) nor any ablation that isolates the balanced-growth rule from seed placement or parameter tuning on this specific collection.
  3. [§3] §3 (Method): the balanced-growth stopping criterion is described only at a high level; the precise rule that decides when to stop expanding a region (intensity threshold schedule, neighborhood size, or balance metric) is not formalized in an equation or pseudocode, preventing independent reproduction or verification that the reported gains are attributable to this mechanism rather than dataset-specific tuning.
minor comments (3)
  1. [Abstract] The abstract and §1 cite 'well-known methods from the literature' without naming them or providing citations in the abstract; the comparison methods should be listed with references already in the abstract for clarity.
  2. [§5] Processing-time claims are stated as 'equivalent to the state-of-the-art' but no absolute times, hardware specification, or per-vertebra timing table is supplied, making the efficiency claim difficult to evaluate.
  3. [§4] The dataset description mentions 102 crushed and 89 normal vertebrae but gives no information on MRI field strength, sequence parameters, or acquisition site, which are standard for reproducibility in medical-image papers.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We appreciate the referee's detailed feedback on our manuscript. We address each of the major comments below, indicating the revisions we plan to make to improve the clarity and rigor of the presentation.

read point-by-point responses

  1. Referee: [§4 and §5] §4 (Experimental Setup) and §5 (Results): the central claim that BGrowth 'significantly outperforms' the compared methods rests on aggregate accuracy figures alone. No definition is given for the accuracy measure (Dice, Jaccard, pixel-wise, or Hausdorff), no per-method means ± std are reported, and no paired statistical test (Wilcoxon signed-rank or t-test) across the 191 cases is performed. This directly undermines the 'significantly outperforms' assertion.

    Authors: We agree with the referee that the accuracy measure requires explicit definition and that reporting means ± std along with statistical tests would strengthen the claims. In the revised manuscript, we will explicitly define the accuracy measure used, report mean and standard deviation for each method across the 191 vertebrae, and include results from a paired statistical test such as the Wilcoxon signed-rank test to assess the significance of the performance differences. These additions will be made to sections 4 and 5. revision: yes

  2. Referee: [§5] §5, Table 2 (or equivalent comparison table): the manuscript states that BGrowth remains best even with 'rough (sloppy) manual annotation,' yet provides no quantitative characterization of seed-point variability (number of seeds, placement tolerance, or inter-observer agreement) nor any ablation that isolates the balanced-growth rule from seed placement or parameter tuning on this specific collection.

    Authors: The referee correctly notes the lack of quantitative details on seed variability and an ablation study. We will add a new subsection or table characterizing the seed points used (e.g., number of seeds per vertebra and placement variability) and include an ablation study comparing BGrowth with and without the balanced-growth rule. This will help isolate the contribution of the proposed mechanism. However, the existing experiments already demonstrate robustness to rough seeds compared to other methods. revision: yes

  3. Referee: [§3] §3 (Method): the balanced-growth stopping criterion is described only at a high level; the precise rule that decides when to stop expanding a region (intensity threshold schedule, neighborhood size, or balance metric) is not formalized in an equation or pseudocode, preventing independent reproduction or verification that the reported gains are attributable to this mechanism rather than dataset-specific tuning.

    Authors: We acknowledge that the description of the balanced-growth stopping criterion in section 3 is at a high level. In the revision, we will provide a formal mathematical definition of the stopping criterion, including the intensity threshold schedule and balance metric, along with pseudocode for the algorithm to enable independent reproduction and verification. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance comparison on held-out vertebrae dataset

full rationale

The paper introduces a semi-automatic segmentation algorithm (BGrowth) and evaluates it via accuracy on a fixed collection of 191 vertebrae against literature baselines. No equations, fitted parameters renamed as predictions, self-definitional constructs, or load-bearing self-citations appear in the provided text. The central claim rests on direct experimental measurement rather than any derivation that reduces to its own inputs by construction. This matches the default case of a self-contained empirical study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are described. The central claim rests on empirical performance claims whose supporting details are absent.

pith-pipeline@v0.9.0 · 5705 in / 1073 out tokens · 27844 ms · 2026-05-25T19:37:24.327263+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

30 extracted references · 30 canonical work pages

  1. [1]

    MEGALE et al

    Megale et al. MEGALE et al. 2017 MEGALE, R. Z. et al. M anagement of vertebral compression fracture in general practice: B E A C H program . PLoS ONE, v. 12, n. 5, p. e0176351, 2017

    work page 2017

  2. [2]

    ZHAO et al

    Zhao et al. ZHAO et al. 2018 ZHAO, W. T. et al. B iomechanical effects of different vertebral heights after augmentation of osteoporotic vertebral compression fracture: a three-dimensional finite element analysis . J Orthop Surg Res, v. 13, n. 1, p. 32, Feb 2018

    work page 2018

  3. [3]

    BARBIERI et al

    Barbieri et al. BARBIERI et al. 2015 BARBIERI, P. D. et al. Vertebral body segmentation of spine mr images using superpixels. In: JUNIOR, C. T. et al. (Ed.). 28th IEEE International Symposium on Computer-Based Medical Systems. S\ ao Carlos and Ribeir\ ao Preto, Brazil: Conference Publishing Services (CPS), 2015. p. 44--49. ISSN 1063-7125

    work page 2015

  4. [4]

    Advances in mr imaging of vertebral collapse

    Tehranzadeh e Tao TEHRANZADEH; TAO 2004 TEHRANZADEH, J.; TAO, C. Advances in mr imaging of vertebral collapse. Seminars in Ultrasound, CT and MRI, v. 25, n. 6, p. 440 -- 460, 2004. ISSN 0887-2171. Imaging of Low Back Pain

    work page 2004

  5. [5]

    R.; THOMA, G

    Antani, Long e Thoma ANTANI; LONG; THOMA 2008 ANTANI, S.; LONG, L. R.; THOMA, G. R. Bridging the gap: Enabling cbir in medical applications. In: 2008 21st IEEE International Symposium on Computer-Based Medical Systems. [S.l.: s.n.], 2008. p. 4--6. ISSN 1063-7125

    work page 2008

  6. [6]

    PAHOLPAK et al

    Paholpak et al. PAHOLPAK et al. 2018 PAHOLPAK, P. et al. Evaluation of foraminal cross-sectional area in lumbar spondylolisthesis using kinematic mri. European Journal of Orthopaedic Surgery & Traumatology , Jul 2018. ISSN 1432-1068

    work page 2018

  7. [7]

    Malignant and benign compression fractures: differentiation and diagnostic pitfalls on mri

    Uetani, Hashmi e Hayashi UETANI; HASHMI; HAYASHI 2004 UETANI, M.; HASHMI, R.; HAYASHI, K. Malignant and benign compression fractures: differentiation and diagnostic pitfalls on mri. Clinical Radiology, v. 59, n. 2, p. 124 -- 131, 2004. ISSN 0009-9260

    work page 2004

  8. [8]

    V ertebral body segmentation with G row C ut: I nitial experience, workflow and practical application

    Egger, Nimsky e Chen EGGER; NIMSKY; CHEN 2017 EGGER, J.; NIMSKY, C.; CHEN, X. V ertebral body segmentation with G row C ut: I nitial experience, workflow and practical application . SAGE Open Med, v. 5, p. 2050312117740984, 2017

    work page 2017

  9. [9]

    CASTI et al

    Casti et al. CASTI et al. 2017 CASTI, P. et al. C ooperative strategy for a dynamic ensemble of classification models in clinical applications: the case of M R I vertebral compression fractures. International Journal of Computer Assisted Radiology and Surgery, v. 12, n. 11, p. 1971--1983, Nov 2017

    work page 2017

  10. [10]

    DZENAN et al

    Dzenan et al. DZENAN et al. 2014 DZENAN, Z. et al. Robust detection and segmentation for diagnosis of vertebral diseases using routine mr images. Computer Graphics Forum, v. 33, n. 6, p. 190--204, 2014

    work page 2014

  11. [11]

    EL-BAZ et al

    El-Baz et al. EL-BAZ et al. 2013 EL-BAZ, A. et al. Computer-aided diagnosis systems for lung cancer: Challenges and methodologies. v. 2013, p. 942353, 2013

    work page 2013

  12. [12]

    SOLIMAN et al

    Soliman et al. SOLIMAN et al. 2018 SOLIMAN, A. et al. A novel cnn segmentation framework based on using new shape and appearance features. In: 2018 25th IEEE International Conference on Image Processing (ICIP). [S.l.: s.n.], 2018. p. 3488--3492. ISSN 2381-8549

    work page 2018

  13. [13]

    HAFRI et al

    Hafri et al. HAFRI et al. 2016a HAFRI, M. et al. Fuzzy energy based active contours model for hr-pqct cortical bone segmentation. In: 2016 IEEE International Conference on Image Processing (ICIP). [S.l.: s.n.], 2016. p. 4334--4338. ISSN 2381-8549

    work page 2016

  14. [14]

    HAFRI et al

    Hafri et al. HAFRI et al. 2016b HAFRI, M. et al. Dual active contours model for hr-pqct cortical bone segmentation. In: 2016 23rd International Conference on Pattern Recognition (ICPR). [S.l.: s.n.], 2016. p. 2270--2275

    work page 2016

  15. [15]

    GrowCut - interactive multi-label N-D image segmentation by cellular automata

    Vezhnevets e Konouchine VEZHNEVETS; KONOUCHINE 2005 VEZHNEVETS, V.; KONOUCHINE, V. GrowCut - interactive multi-label N-D image segmentation by cellular automata. International Conference on Computer Graphics and Vision - GraphiCon , v. 1, Nov 2005

    work page 2005

  16. [16]

    ZHU et al

    Zhu et al. ZHU et al. 2014 ZHU, L. et al. An effective interactive medical image segmentation method using fast growcut. In: . [S.l.: s.n.], 2014. v. 17, n. WS

    work page 2014

  17. [17]

    JUNIOR et al

    Junior et al. JUNIOR et al. 2018 JUNIOR, J. R. F. et al. Radiomics-based features for pattern recognition of lung cancer histopathology and metastases. Computer Methods and Programs in Biomedicine, v. 159, p. 23 -- 30, 2018. ISSN 0169-2607

    work page 2018

  18. [18]

    Stochastic relaxation, gibbs distributions, and the bayesian restoration of images

    Geman e Geman GEMAN; GEMAN 1984 GEMAN, S.; GEMAN, D. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-6, n. 6, p. 721--741, Nov 1984. ISSN 0162-8828

    work page 1984

  19. [19]

    GrabCut : Interactive foreground extraction using iterated graph cuts

    Rother, Kolmogorov e Blake ROTHER; KOLMOGOROV; BLAKE 2004 ROTHER, C.; KOLMOGOROV, V.; BLAKE, A. GrabCut : Interactive foreground extraction using iterated graph cuts. ACM Trans. Graph., ACM, New York, USA, v. 23, n. 3, p. 309--314, ago. 2004. ISSN 0730-0301

    work page 2004

  20. [20]

    LI et al

    Li et al. LI et al. 2004 LI, Y. et al. Lazy snapping. ACM Trans. Graph., ACM, New York, USA, v. 23, n. 3, p. 303--308, ago. 2004. ISSN 0730-0301

    work page 2004

  21. [21]

    YONG et al

    Yong et al. YONG et al. 2017 YONG, Z. et al. Grabcut image segmentation algorithm based on structure tensor. The Journal of China Universities of Posts and Telecommunications, v. 24, n. 2, p. 38 -- 47, 2017. ISSN 1005-8885

    work page 2017

  22. [22]

    LU et al

    Lu et al. LU et al. 2017 LU, Y.-W. et al. Segmentation method for medical image based on improved grabcut. International Journal of Imaging Systems and Technology, v. 27, n. 4, p. 383--390, 2017

    work page 2017

  23. [23]

    Automatic grabcut based lung extraction from endoscopic images with an initial boundary

    Wu, Nakao e Matsuda WU; NAKAO; MATSUDA 2016 WU, S.; NAKAO, M.; MATSUDA, T. Automatic grabcut based lung extraction from endoscopic images with an initial boundary. In: 2016 IEEE 13th International Conference on Signal Processing (ICSP). [S.l.: s.n.], 2016. p. 1374--1378. ISSN 2164-5221

    work page 2016

  24. [24]

    The distribution of the flora in the alpine zone

    Jaccard JACCARD 1912 JACCARD, P. The distribution of the flora in the alpine zone. New Phytologist, v. 11, n. 2, p. 37--50, fev. 1912

    work page 1912

  25. [25]

    A Method of Establishing Groups of Equal Amplitude in Plant Sociology Based on Similarity of Species Content and Its Application to Analyses of the Vegetation on Danish Commons

    S rensen S RENSEN 1948 S RENSEN, T. A Method of Establishing Groups of Equal Amplitude in Plant Sociology Based on Similarity of Species Content and Its Application to Analyses of the Vegetation on Danish Commons. [S.l.]: I kommission hos E. Munksgaard, 1948. (Biologiske skrifter)

    work page 1948

  26. [26]

    RAMOS et al

    Ramos et al. RAMOS et al. 2017 RAMOS, J. S. et al. How to speed up outliers removal in image matching. Pattern Recognition Letters, 2017. ISSN 0167-8655

    work page 2017

  27. [27]

    S.; WATANABE, C

    Ramos, Watanabe e Traina RAMOS; WATANABE; TRAINA 2016 RAMOS, J. S.; WATANABE, C. Y. V.; TRAINA, A. J. M. Fomp: A novel preprocessing technique to speed-up the outlier removal from matched points. In: 2016 29th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI). [S.l.: s.n.], 2016. p. 225--232

    work page 2016

  28. [28]

    A probabilistic interpretation of precision, recall and f-score, with implication for evaluation

    Goutte e Gaussier GOUTTE; GAUSSIER 2005 GOUTTE, C.; GAUSSIER, E. A probabilistic interpretation of precision, recall and f-score, with implication for evaluation. In: Proceedings of the 27th European Conference on Advances in Information Retrieval Research. Berlin, Heidelberg: [s.n.], 2005. (ECIR'05), p. 345--359. ISBN 3-540-25295-9, 978-3-540-25295-5

    work page 2005

  29. [29]

    Massey MASSEY 1951 MASSEY, F. J. The K olmogorov- S mirnov test for goodness of fit. Journal of the American Statistical Association, American Statistical Association, v. 46, n. 253, p. 68--78, 1951

    work page 1951

  30. [30]

    Critical values and probability levels for the wilcoxon rank sum test and the wilcoxon signed rank test

    Wilcoxon, Katti e Wilcox WILCOXON; KATTI; WILCOX 1970 WILCOXON, F.; KATTI, S.; WILCOX, R. Critical values and probability levels for the wilcoxon rank sum test and the wilcoxon signed rank test. Selected Tables in Mathematical Statistics, v. 1, p. 171--259, 1970

    work page 1970