Deep-Learning for Tidemark Segmentation in Human Osteochondral Tissues Imaged with Micro-computed Tomography

Aleksei Tiulpin; Heikki J. Nieminen; Mikko Finnil\"a; Petri Lehenkari; Simo Saarakkala

arxiv: 1907.05089 · v1 · pith:FZGEW76Znew · submitted 2019-07-11 · 📡 eess.IV · cs.CV

Deep-Learning for Tidemark Segmentation in Human Osteochondral Tissues Imaged with Micro-computed Tomography

Aleksei Tiulpin , Mikko Finnil\"a , Petri Lehenkari , Heikki J. Nieminen , Simo Saarakkala This is my paper

Pith reviewed 2026-05-24 23:16 UTC · model grok-4.3

classification 📡 eess.IV cs.CV

keywords tidemark segmentationU-Netmicro-computed tomographyosteochondral tissueosteoarthritisPTA stainingdeep learningimage segmentation

0 comments

The pith

U-Net trained on binary cross-entropy and soft Jaccard loss segments the tidemark automatically in PTA-stained micro-CT osteochondral images.

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

The paper shows the first deep-learning approach for segmenting the tidemark, or calcified cartilage interface, in three-dimensional micro-computed tomography scans of human articular cartilage after phosphotungstic acid staining. The standard imaging protocol loses contrast exactly at this interface, yet precise segmentation supports better understanding of osteoarthritis development and early-stage tissue assessment. The method applies a U-Net architecture trained with a mix of binary cross-entropy and soft Jaccard loss. Cross-validation across 35 samples produces intersection-over-union scores that rise steadily from 0.59 within a 15 micrometer tolerance zone to 0.86 within a 75 micrometer zone. Codes, the full dataset, and segmentation masks are released publicly to encourage further work on biomedical image segmentation.

Core claim

A U-Net model trained using a combination of binary cross-entropy and soft Jaccard loss performs fully automatic segmentation of the tidemark in PTA-stained osteochondral tissues imaged with micro-computed tomography, achieving intersection over union values of 0.59, 0.70, 0.79, 0.83, and 0.86 within 15 μm, 30 μm, 45 μm, 60 μm, and 75 μm padded zones around the tidemark on cross-validation.

What carries the argument

U-Net neural network trained with binary cross-entropy combined with soft Jaccard loss for binary segmentation of the calcified cartilage interface.

Load-bearing premise

The 35 PTA-stained human samples used for training and cross-validation represent the full range of tidemark appearance, staining variability, and scanner settings encountered in osteoarthritis studies.

What would settle it

A test set of new PTA-stained samples acquired under different staining conditions or scanner settings that produces intersection-over-union scores markedly below the reported cross-validation numbers at each tolerance zone.

Figures

Figures reproduced from arXiv: 1907.05089 by Aleksei Tiulpin, Heikki J. Nieminen, Mikko Finnil\"a, Petri Lehenkari, Simo Saarakkala.

**Figure 1.** Figure 1: Examples of the slices from ◦ = 2mm human osteochondral plugs imaged with contrast-enhanced µCT. a) a typical sample showing the loss of the contrast at the CCI. b) a typical non-enhancing region (void) which occurs with some samples. PTA. However, both of these options are time consuming and could be avoided with the help of Machine Learning. In clinical OA research Machine Learning is has been applied to… view at source ↗

**Figure 2.** Figure 2: Data acquisition pipeline: from sample preparation to imaging. 2.2 Data pre-processing Our imaging protocol allowed to obtain the 3D volumes of human cartilage and the mask annotations for the underlying mineralized tissues. The original size of the reconstructed samples ranges from 756×756 to 1008×1008 pixels in width and 884 to 2067 pixels in height (including the empty space around the sample). To harm… view at source ↗

**Figure 3.** Figure 3: Data processing pipeline. Here, we co-registered CA4+ and PTA samples and obtained the segmentation masks for hard tissues. These masks were used in training of our segmentation model [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Figure 4: Median values of performance metrics for different levels of padding around the tidemark. Here, subplots (a), (b) and (c) show the performance for IoU, Dice and Volumetric similarity scores, respectively [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

read the original abstract

Three-dimensional (3D) semi-quantitative grading of pathological features in articular cartilage (AC) offers significant improvements in basic research of osteoarthritis (OA). We have earlier developed the 3D protocol for imaging of AC and its structures which includes staining of the sample with a contrast agent (phosphotungstic acid, PTA) and a consequent scanning with micro-computed tomography. Such a protocol was designed to provide X-ray attenuation contrast to visualize AC structure. However, at the same time, this protocol has one major disadvantage: the loss of contrast at the tidemark (calcified cartilage interface, CCI). An accurate segmentation of CCI can be very important for understanding the etiology of OA and ex-vivo evaluation of tidemark condition at early OA stages. In this paper, we present the first application of Deep Learning to PTA-stained osteochondral samples that allows to perform tidemark segmentation in a fully-automatic manner. Our method is based on U-Net trained using a combination of binary cross-entropy and soft Jaccard loss. On cross-validation, this approach yielded intersection over the union of 0.59, 0.70, 0.79, 0.83 and 0.86 within 15 {\mu}m, 30 {\mu}m, 45 {\mu}m, 60 {\mu}m and 75 {\mu}m padded zones around the tidemark, respectively. Our codes and the dataset that consisted of 35 PTA-stained human AC samples are made publicly available together with the segmentation masks to facilitate the development of biomedical image segmentation methods.

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.

Desk Editor's Note private letter to a colleague

Straightforward first DL application to tidemark segmentation with public 35-sample release, but generalizability is the open question.

read the letter

This paper shows that a standard U-Net trained with binary cross-entropy plus soft Jaccard loss can segment the tidemark in PTA-stained osteochondral micro-CT scans. The main contribution is the domain application itself—the first reported use of deep learning for this exact task—plus the release of code, the 35-sample dataset, and the masks. That release is the part that actually moves the field forward for anyone who wants to build on it or test extensions. The evaluation reports IoU at five tolerance distances on cross-validation folds, which gives a clearer picture of practical utility than a single number would. The numbers rise from 0.59 at 15 μm to 0.86 at 75 μm, which is reasonable for the imaging protocol described. The architecture and loss are not new, so the work stands or falls on how well the application is executed and documented. The central empirical claim is supported by the held-out folds, with no obvious circularity in the metrics. The soft spot is the dataset size and homogeneity: 35 samples from one staining and scanning protocol, with no breakdown by patient, OA grade, or scanner settings. That makes it hard to judge whether the performance will hold for the wider range of tidemark appearances and technical variability that OA studies actually encounter. No inter-rater variability is reported either, so there is no external anchor for the IoU values. This is useful for labs already working with PTA micro-CT who need an automated starting point for 3D grading or who want to expand the public set. It is less useful for anyone claiming broad clinical or multi-center applicability without further validation. The thinking is clear and the execution is honest within its scope. Send it to peer review; the dataset release alone makes it worth referee time even if the reviewers push on the limited cohort.

Referee Report

1 major / 3 minor

Summary. The manuscript presents the first application of deep learning for automatic tidemark (calcified cartilage interface) segmentation in PTA-stained human osteochondral samples imaged via micro-CT. It employs a U-Net trained with binary cross-entropy combined with soft Jaccard loss and reports 5-fold cross-validation IoU scores of 0.59, 0.70, 0.79, 0.83, and 0.86 at tolerance distances of 15, 30, 45, 60, and 75 μm on a released dataset of 35 samples, with code and data made public.

Significance. If the performance holds under broader conditions, the approach could accelerate 3D semi-quantitative grading of osteoarthritis features by replacing manual tidemark delineation. The public release of the 35-sample dataset, segmentation masks, and code is a clear strength that supports reproducibility and community development of biomedical segmentation methods.

major comments (1)

[Abstract, results paragraph on cross-validation performance] Abstract and results paragraph on cross-validation performance: the IoU values are obtained from a single cohort of 35 PTA-stained samples imaged under one staining/scanning protocol, with no reported stratification by patient, OA grade, or scanner parameters. This leaves the assumption that the cohort captures the full range of tidemark appearance, staining variability, and technical conditions untested, which is load-bearing for the claim that the method enables fully-automatic segmentation for OA studies in general.

minor comments (3)

[Methods] Methods section: full training details (hyperparameters, exact U-Net depth, data augmentation strategy, and fold definitions) are referenced only at a high level; explicit listing would strengthen reproducibility claims.
[Results] Results: no comparison to inter-rater variability among human annotators or to simpler baselines (e.g., intensity thresholding) is provided, which would help contextualize the reported IoU values at each tolerance distance.
Figure captions and text: the definition of the 'padded zones' around the tidemark (how the tolerance distance is applied to the ground-truth mask) could be stated more explicitly to avoid ambiguity in interpreting the multi-distance IoU metrics.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. We address the single major comment below.

read point-by-point responses

Referee: [Abstract, results paragraph on cross-validation performance] Abstract and results paragraph on cross-validation performance: the IoU values are obtained from a single cohort of 35 PTA-stained samples imaged under one staining/scanning protocol, with no reported stratification by patient, OA grade, or scanner parameters. This leaves the assumption that the cohort captures the full range of tidemark appearance, staining variability, and technical conditions untested, which is load-bearing for the claim that the method enables fully-automatic segmentation for OA studies in general.

Authors: We agree that the reported IoU values derive from a single cohort of 35 samples acquired under one consistent PTA-staining and micro-CT protocol, with no stratification by patient, OA grade, or scanner parameters described in the manuscript. This restricts the strength of any claim to broad generalizability. In the revised version we will (i) rephrase the abstract and results paragraph to state explicitly that performance is reported for this specific cohort and acquisition protocol, and (ii) insert a dedicated limitations paragraph that acknowledges the untested variability across staining conditions, scanner parameters, and patient populations and calls for multi-cohort validation. The public release of the dataset and code is intended to support exactly such extensions. revision: yes

Circularity Check

0 steps flagged

No circularity: performance metrics obtained from standard cross-validation on held-out folds

full rationale

The paper applies a standard U-Net architecture trained with binary cross-entropy plus soft Jaccard loss to segment the tidemark in PTA-stained micro-CT volumes. The reported IoU values (0.59–0.86 across distance-padded zones) are computed directly from held-out folds of the 35-sample dataset; no equation, loss term, or parameter is defined in terms of the target metric itself. No self-citation chain is invoked to justify uniqueness or to substitute for an external result, and the method description contains no ansatz, renaming of known patterns, or fitted-input-called-prediction structure. The evaluation is therefore self-contained against the released data splits and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the empirical performance of a standard U-Net trained on the released 35-sample dataset; no new free parameters, invented physical entities, or non-standard mathematical axioms are introduced beyond the usual assumptions of supervised segmentation.

axioms (1)

domain assumption The distribution of tidemark appearance and imaging artifacts in the 35 training samples is representative of the broader population of human osteochondral tissues.
Cross-validation performance is reported without external test sets or explicit discussion of domain shift.

pith-pipeline@v0.9.0 · 5846 in / 1372 out tokens · 23930 ms · 2026-05-24T23:16:54.079593+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Automatic Grading of Individual Knee Osteoarthritis Features in Plain Radiographs using Deep Convolutional Neural Networks
eess.IV 2019-07 unverdicted novelty 4.0

Multi-task CNN ensemble achieves kappa 0.82 for KL and high kappas for OARSI features on independent test set, with AUC 0.98 for OA detection.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · cited by 1 Pith paper · 3 internal anchors

[1]

Computers in biology and medicine 95, 24–33 (2018)

Abidin, A.Z., Deng, B., DSouza, A.M., Nagarajan, M.B., Coan, P., Wism¨ uller, A.: Deep transfer learning for characterizing chondrocyte patterns in phase contrast x-ray computed tomography images of the human patellar cartilage. Computers in biology and medicine 95, 24–33 (2018)

work page 2018
[2]

In: 2016 23rd International Conference on Pattern Recognition (ICPR)

Antony, J., McGuinness, K., O’Connor, N.E., Moran, K.: Quantifying radiographic knee osteoarthritis severity using deep convolutional neural networks. In: 2016 23rd International Conference on Pattern Recognition (ICPR). pp. 1195–1200. IEEE (2016)

work page 2016
[3]

Bradski, G.: The OpenCV Library. Dr. Dobb’s Journal of Software Tools (2000) 6 Tiulpin A. et al

work page 2000
[4]

The Lancet 386(9991), 376–387 (2015)

Glyn-Jones, S., Palmer, A., Agricola, R., Price, A., Vincent, T., Weinans, H., Carr, A.: Osteoarthritis. The Lancet 386(9991), 376–387 (2015)

work page 2015
[5]

TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation

Iglovikov, V., Shvets, A.: Ternausnet: U-net with vgg11 encoder pre-trained on imagenet for image segmentation. arXiv preprint arXiv:1801.05746 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018
[6]

Frontiers in Physics 5, 38 (2017)

Karhula, S.S., Finnil¨ a, M.A., Freedman, J.D., Kauppinen, S., Valkealahti, M., Lehenkari, P., Pritzker, K.P., Nieminen, H.J., Snyder, B.D., Grinstaﬀ, M.W., et al.: Micro-scale distribution of ca4+ in ex vivo human articular cartilage detected with contrast-enhanced micro-computed tomography imaging. Frontiers in Physics 5, 38 (2017)

work page 2017
[7]

PloS one 12(1), e0171075 (2017)

Karhula, S.S., Finnil¨ a, M.A., Lammi, M.J., Yl¨ arinne, J.H., Kauppinen, S., Rieppo, L., Pritzker, K.P., Nieminen, H.J., Saarakkala, S.: Eﬀects of articular cartilage constituents on phosphotungstic acid enhanced micro-computed tomography. PloS one 12(1), e0171075 (2017)

work page 2017
[8]

Osteoarthritis and cartilage 27(1), 172–180 (2019)

Kauppinen, S., Karhula, S., Thevenot, J., Ylitalo, T., Rieppo, L., Kestil¨ a, I., Haapea, M., Hadjab, I., Finnil¨ a, M., Quenneville, E., et al.: 3d morphometric analysis of calciﬁed cartilage properties using micro-computed tomography. Osteoarthritis and cartilage 27(1), 172–180 (2019)

work page 2019
[9]

Adam: A Method for Stochastic Optimization

Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[10]

Osteoarthritis and cartilage 25(10), 1680–1689 (2017)

Nieminen, H., Gahunia, H., Pritzker, K., Ylitalo, T., Rieppo, L., Karhula, S., Lehenkari, P., Hæggstr¨ om, E., Saarakkala, S.: 3d histopathological grading of osteochondral tissue using contrast-enhanced micro- computed tomography. Osteoarthritis and cartilage 25(10), 1680–1689 (2017)

work page 2017
[11]

Osteoarthritis and cartilage 23(9), 1613– 1621 (2015)

Nieminen, H., Ylitalo, T., Karhula, S., Suuronen, J.P., Kauppinen, S., Serimaa, R., Hæggstr¨ om, E., Pritzker, K., Valkealahti, M., Lehenkari, P., et al.: Determining collagen distribution in articular car- tilage using contrast-enhanced micro-computed tomography. Osteoarthritis and cartilage 23(9), 1613– 1621 (2015)

work page 2015
[12]

Radiology 288(1), 177–185 (2018)

Norman, B., Pedoia, V., Majumdar, S.: Use of 2d u-net convolutional neural networks for automated cartilage and meniscus segmentation of knee mr imaging data to determine relaxometry and morphom- etry. Radiology 288(1), 177–185 (2018)

work page 2018
[13]

Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic diﬀerentiation in pytorch (2017)

work page 2017
[14]

Journal of microscopy 250(1), 21–31 (2013)

Pauwels, E., Van Loo, D., Cornillie, P., Brabant, L., Van Hoorebeke, L.: An exploratory study of contrast agents for soft tissue visualization by means of high resolution x-ray computed tomography imaging. Journal of microscopy 250(1), 21–31 (2013)

work page 2013
[15]

Magnetic Resonance Materials in Physics, Biology and Medicine 29(2), 207–221 (2016)

Pedoia, V., Majumdar, S., Link, T.M.: Segmentation of joint and musculoskeletal tissue in the study of arthritis. Magnetic Resonance Materials in Physics, Biology and Medicine 29(2), 207–221 (2016)

work page 2016
[16]

In: International Conference on Medical image computing and computer-assisted intervention

Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmen- tation. In: International Conference on Medical image computing and computer-assisted intervention. pp. 234–241. Springer (2015)

work page 2015
[17]

Annual review of biomedical engineering 19, 221–248 (2017)

Shen, D., Wu, G., Suk, H.I.: Deep learning in medical image analysis. Annual review of biomedical engineering 19, 221–248 (2017)

work page 2017
[18]

https://github.com/MIPT-Oulu/solt (2019)

Tiulpin, A.: Solt: Streaming over lightweight transformations. https://github.com/MIPT-Oulu/solt (2019)

work page 2019
[19]

Multimodal Machine Learning-based Knee Osteoarthritis Progression Prediction from Plain Radiographs and Clinical Data

Tiulpin, A., Klein, S., Bierma-Zeinstra, S., Thevenot, J., Rahtu, E., van Meurs, J., Oei, E.H., Saarakkala, S.: Multimodal machine learning-based knee osteoarthritis progression prediction from plain radiographs and clinical data. arXiv preprint arXiv:1904.06236 (2019)

work page internal anchor Pith review Pith/arXiv arXiv 1904
[20]

Scientiﬁc reports 8(1), 1727 (2018)

Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., Saarakkala, S.: Automatic knee osteoarthritis diagnosis from plain radiographs: a deep learning-based approach. Scientiﬁc reports 8(1), 1727 (2018)

work page 2018
[21]

In: Scandinavian Conference on Image Analysis

Tiulpin, A., Thevenot, J., Rahtu, E., Saarakkala, S.: A novel method for automatic localization of joint area on knee plain radiographs. In: Scandinavian Conference on Image Analysis. pp. 290–301. Springer (2017)

work page 2017

[1] [1]

Computers in biology and medicine 95, 24–33 (2018)

Abidin, A.Z., Deng, B., DSouza, A.M., Nagarajan, M.B., Coan, P., Wism¨ uller, A.: Deep transfer learning for characterizing chondrocyte patterns in phase contrast x-ray computed tomography images of the human patellar cartilage. Computers in biology and medicine 95, 24–33 (2018)

work page 2018

[2] [2]

In: 2016 23rd International Conference on Pattern Recognition (ICPR)

Antony, J., McGuinness, K., O’Connor, N.E., Moran, K.: Quantifying radiographic knee osteoarthritis severity using deep convolutional neural networks. In: 2016 23rd International Conference on Pattern Recognition (ICPR). pp. 1195–1200. IEEE (2016)

work page 2016

[3] [3]

Bradski, G.: The OpenCV Library. Dr. Dobb’s Journal of Software Tools (2000) 6 Tiulpin A. et al

work page 2000

[4] [4]

The Lancet 386(9991), 376–387 (2015)

Glyn-Jones, S., Palmer, A., Agricola, R., Price, A., Vincent, T., Weinans, H., Carr, A.: Osteoarthritis. The Lancet 386(9991), 376–387 (2015)

work page 2015

[5] [5]

TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation

Iglovikov, V., Shvets, A.: Ternausnet: U-net with vgg11 encoder pre-trained on imagenet for image segmentation. arXiv preprint arXiv:1801.05746 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018

[6] [6]

Frontiers in Physics 5, 38 (2017)

Karhula, S.S., Finnil¨ a, M.A., Freedman, J.D., Kauppinen, S., Valkealahti, M., Lehenkari, P., Pritzker, K.P., Nieminen, H.J., Snyder, B.D., Grinstaﬀ, M.W., et al.: Micro-scale distribution of ca4+ in ex vivo human articular cartilage detected with contrast-enhanced micro-computed tomography imaging. Frontiers in Physics 5, 38 (2017)

work page 2017

[7] [7]

PloS one 12(1), e0171075 (2017)

Karhula, S.S., Finnil¨ a, M.A., Lammi, M.J., Yl¨ arinne, J.H., Kauppinen, S., Rieppo, L., Pritzker, K.P., Nieminen, H.J., Saarakkala, S.: Eﬀects of articular cartilage constituents on phosphotungstic acid enhanced micro-computed tomography. PloS one 12(1), e0171075 (2017)

work page 2017

[8] [8]

Osteoarthritis and cartilage 27(1), 172–180 (2019)

Kauppinen, S., Karhula, S., Thevenot, J., Ylitalo, T., Rieppo, L., Kestil¨ a, I., Haapea, M., Hadjab, I., Finnil¨ a, M., Quenneville, E., et al.: 3d morphometric analysis of calciﬁed cartilage properties using micro-computed tomography. Osteoarthritis and cartilage 27(1), 172–180 (2019)

work page 2019

[9] [9]

Adam: A Method for Stochastic Optimization

Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014

[10] [10]

Osteoarthritis and cartilage 25(10), 1680–1689 (2017)

Nieminen, H., Gahunia, H., Pritzker, K., Ylitalo, T., Rieppo, L., Karhula, S., Lehenkari, P., Hæggstr¨ om, E., Saarakkala, S.: 3d histopathological grading of osteochondral tissue using contrast-enhanced micro- computed tomography. Osteoarthritis and cartilage 25(10), 1680–1689 (2017)

work page 2017

[11] [11]

Osteoarthritis and cartilage 23(9), 1613– 1621 (2015)

Nieminen, H., Ylitalo, T., Karhula, S., Suuronen, J.P., Kauppinen, S., Serimaa, R., Hæggstr¨ om, E., Pritzker, K., Valkealahti, M., Lehenkari, P., et al.: Determining collagen distribution in articular car- tilage using contrast-enhanced micro-computed tomography. Osteoarthritis and cartilage 23(9), 1613– 1621 (2015)

work page 2015

[12] [12]

Radiology 288(1), 177–185 (2018)

Norman, B., Pedoia, V., Majumdar, S.: Use of 2d u-net convolutional neural networks for automated cartilage and meniscus segmentation of knee mr imaging data to determine relaxometry and morphom- etry. Radiology 288(1), 177–185 (2018)

work page 2018

[13] [13]

Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic diﬀerentiation in pytorch (2017)

work page 2017

[14] [14]

Journal of microscopy 250(1), 21–31 (2013)

Pauwels, E., Van Loo, D., Cornillie, P., Brabant, L., Van Hoorebeke, L.: An exploratory study of contrast agents for soft tissue visualization by means of high resolution x-ray computed tomography imaging. Journal of microscopy 250(1), 21–31 (2013)

work page 2013

[15] [15]

Magnetic Resonance Materials in Physics, Biology and Medicine 29(2), 207–221 (2016)

Pedoia, V., Majumdar, S., Link, T.M.: Segmentation of joint and musculoskeletal tissue in the study of arthritis. Magnetic Resonance Materials in Physics, Biology and Medicine 29(2), 207–221 (2016)

work page 2016

[16] [16]

In: International Conference on Medical image computing and computer-assisted intervention

Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmen- tation. In: International Conference on Medical image computing and computer-assisted intervention. pp. 234–241. Springer (2015)

work page 2015

[17] [17]

Annual review of biomedical engineering 19, 221–248 (2017)

Shen, D., Wu, G., Suk, H.I.: Deep learning in medical image analysis. Annual review of biomedical engineering 19, 221–248 (2017)

work page 2017

[18] [18]

https://github.com/MIPT-Oulu/solt (2019)

Tiulpin, A.: Solt: Streaming over lightweight transformations. https://github.com/MIPT-Oulu/solt (2019)

work page 2019

[19] [19]

Multimodal Machine Learning-based Knee Osteoarthritis Progression Prediction from Plain Radiographs and Clinical Data

Tiulpin, A., Klein, S., Bierma-Zeinstra, S., Thevenot, J., Rahtu, E., van Meurs, J., Oei, E.H., Saarakkala, S.: Multimodal machine learning-based knee osteoarthritis progression prediction from plain radiographs and clinical data. arXiv preprint arXiv:1904.06236 (2019)

work page internal anchor Pith review Pith/arXiv arXiv 1904

[20] [20]

Scientiﬁc reports 8(1), 1727 (2018)

Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., Saarakkala, S.: Automatic knee osteoarthritis diagnosis from plain radiographs: a deep learning-based approach. Scientiﬁc reports 8(1), 1727 (2018)

work page 2018

[21] [21]

In: Scandinavian Conference on Image Analysis

Tiulpin, A., Thevenot, J., Rahtu, E., Saarakkala, S.: A novel method for automatic localization of joint area on knee plain radiographs. In: Scandinavian Conference on Image Analysis. pp. 290–301. Springer (2017)

work page 2017