pith. sign in

arxiv: 2005.10951 · v1 · submitted 2020-05-22 · 🧬 q-bio.QM · q-bio.TO

A machine learning approach to using Quality-of-Life patient scores in guiding prostate radiation therapy dosing

Zhijian Yang , Daniel Olszewski , Chujun He , Giulia Pintea , Jun Lian , Tom Chou , Ronald Chen , Blerta Shtylla This is my paper

Pith reviewed 2026-05-24 15:17 UTC · model grok-4.3

classification 🧬 q-bio.QM q-bio.TO
keywords machine learningprostate cancerradiation therapyquality of lifeconvolutional neural networkrectal dosedosage thresholdsdata augmentation
0
0 comments X

The pith

Radiation to rectal regions correlates with quality-of-life changes after prostate treatment while bladder radiation shows no link.

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

The paper applies machine learning to limited patient data on prostate radiation doses and post-treatment quality-of-life scores. It augments the data through image flipping and curvature-based interpolation, then trains a convolutional autoencoder followed by a convolutional neural network to map dose distributions to score changes. Statistical models identify organ-specific sensitivities and dosage thresholds. A sympathetic reader would care because the results point toward adjusting radiation plans to protect rectal regions and thereby reduce side effects while maintaining cancer control.

Core claim

Analysis of augmented radiation dose maps and patient-reported quality-of-life scores shows no association between bladder radiation and quality-of-life changes, yet identifies associations between radiation to the anterior and posterior rectal regions and such changes, together with estimated dosage thresholds for each organ region.

What carries the argument

Convolutional neural network trained on radiation dose distributions to predict quality-of-life changes, combined with ANOVA and logistic regression to derive organ-specific dosage thresholds.

If this is right

  • Radiation plans can prioritize avoiding threshold doses in anterior and posterior rectal regions to limit quality-of-life impact.
  • Bladder dose can be deprioritized when optimizing plans for quality-of-life preservation.
  • Machine learning models trained on augmented data can supply patient-specific dosage guidance from reported quality-of-life metrics.
  • The same augmentation-plus-network pipeline can be reused for other organs once additional data become available.

Where Pith is reading between the lines

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

  • If the thresholds hold in prospective data, treatment planning software could incorporate direct quality-of-life predictions.
  • The approach could extend to other radiation-sensitive cancers where nearby organs affect daily function.
  • Longer-term follow-up scores might reveal whether the identified rectal thresholds also predict late effects.

Load-bearing premise

Image flipping and curvature-based interpolation generate realistic anatomical and dosimetric variations that preserve the true radiation-to-quality-of-life relationship without systematic bias.

What would settle it

A larger set of unaugmented real patient data in which rectal radiation shows no correlation with quality-of-life score changes would falsify the reported associations and thresholds.

Figures

Figures reproduced from arXiv: 2005.10951 by Blerta Shtylla, Chujun He, Daniel Olszewski, Giulia Pintea, Jun Lian, Ronald Chen, Tom Chou, Zhijian Yang.

Figure 1
Figure 1. Figure 1: Total difference scores in urinary symptoms for patients 1-54. In blue, we mark patients who were classified [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Interpolation of CT images of the bladder using the FM algorithm. The interpolated image (c) is the [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A representative radiation plan. The highest dosage corresponds to the greatest pixel intensity (in white). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Architecture used for the CNN classification model. There are three layers that have convolution, activation, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (A) Model cross-validation. We initialized 5 different models. For each model there is a validation set that moves across the data. First, we trained on the training set, then checked each of the models on the different validation sets, providing a general idea of which model would work the best. (B) Training (orange) and validation (blue) loss for the best model chosen through cross-validation. As shown i… view at source ↗
Figure 6
Figure 6. Figure 6: Convolutional Autoencoder. (A) Schematic of the architecture of the convolutional autoencoder network. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Organ contouring and organ regions. (A) A CT slice of the rectum of patient 40, showing the new [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Accuracy for our trained classification model. The overall accuracy for the bladder and rectum and the [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Logistic model thresholds and corresponding RT dosages for each rectum region. (A) Computed thresholds [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
read the original abstract

Thanks to advancements in diagnosis and treatment, prostate cancer patients have high long-term survival rates. Currently, an important goal is to preserve quality-of-life during and after treatment. The relationship between the radiation a patient receives and the subsequent side effects he experiences is complex and difficult to model or predict. Here, we use machine learning algorithms and statistical models to explore the connection between radiation treatment and post-treatment gastro-urinary function. Since only a limited number of patient datasets are currently available, we used image flipping and curvature-based interpolation methods to generate more data in order to leverage transfer learning. Using interpolated and augmented data, we trained a convolutional autoencoder network to obtain near-optimal starting points for the weights. A convolutional neural network then analyzed the relationship between patient-reported quality-of-life and radiation. We also used analysis of variance and logistic regression to explore organ sensitivity to radiation and develop dosage thresholds for each organ region. Our findings show no connection between the bladder and quality-of-life scores. However, we found a connection between radiation applied to posterior and anterior rectal regions to changes in quality-of-life. Finally, we estimated radiation therapy dosage thresholds for each organ. Our analysis connects machine learning methods with organ sensitivity, thus providing a framework for informing cancer patient care using patient reported quality-of-life metrics.

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 / 0 minor

Summary. The manuscript describes a machine learning pipeline that augments a small cohort of prostate cancer patient datasets via image flipping and curvature-based interpolation, pre-trains a convolutional autoencoder, and then trains a CNN to relate 3D radiation dose distributions to patient-reported quality-of-life (QoL) scores. Statistical analyses (ANOVA and logistic regression) are used to identify organ-region sensitivities and to estimate dosage thresholds; the central claims are the absence of a bladder-QoL association, the presence of associations for anterior and posterior rectal regions, and the derivation of organ-specific dose thresholds.

Significance. If the reported associations and thresholds prove robust to validation, the work would supply a concrete, patient-reported-outcome-driven method for personalizing prostate radiotherapy planning, directly linking dosimetric maps to gastro-urinary function preservation.

major comments (3)
  1. [Abstract] Abstract (data-augmentation paragraph): the claim that flipping and curvature-based interpolation produce realistic anatomical and dosimetric variations is asserted without any supporting check (e.g., Kolmogorov-Smirnov tests on dose histograms, preservation of left-right asymmetry, or comparison of joint dose-QoL distributions before versus after augmentation). Because every downstream CNN weight and every ANOVA/logistic threshold is derived from the augmented set, this unverified step is load-bearing for all three main findings.
  2. [Abstract] Abstract: no quantitative performance metrics, cross-validation scheme, confidence intervals, or external test-set results are supplied for either the CNN or the derived dosage thresholds. Without these, it is impossible to assess whether the reported rectal associations exceed what would be obtained by chance on the original (unaugmented) cohort.
  3. The manuscript provides no explicit statement of the original patient cohort size, the number of synthetic samples generated, or the train/validation/test split ratios; these numbers are required to evaluate whether the transfer-learning step is statistically supported.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful and constructive review. The comments correctly identify several omissions in the submitted manuscript that limit the ability to evaluate the robustness of the augmentation procedure, model performance, and statistical claims. We address each point below and will revise the manuscript to incorporate the requested information and checks.

read point-by-point responses

  1. Referee: [Abstract] Abstract (data-augmentation paragraph): the claim that flipping and curvature-based interpolation produce realistic anatomical and dosimetric variations is asserted without any supporting check (e.g., Kolmogorov-Smirnov tests on dose histograms, preservation of left-right asymmetry, or comparison of joint dose-QoL distributions before versus after augmentation). Because every downstream CNN weight and every ANOVA/logistic threshold is derived from the augmented set, this unverified step is load-bearing for all three main findings.

    Authors: We agree that the current manuscript asserts realism of the augmentation without quantitative verification. In the revised version we will add Kolmogorov-Smirnov tests comparing dose-volume histograms of original versus augmented samples, checks for preservation of left-right asymmetry, and direct comparison of the joint dose-QoL distributions before and after augmentation. These additions will be placed in the Methods and Results sections to support the downstream analyses. revision: yes

  2. Referee: [Abstract] Abstract: no quantitative performance metrics, cross-validation scheme, confidence intervals, or external test-set results are supplied for either the CNN or the derived dosage thresholds. Without these, it is impossible to assess whether the reported rectal associations exceed what would be obtained by chance on the original (unaugmented) cohort.

    Authors: We acknowledge the absence of these metrics. The revised manuscript will report k-fold cross-validation performance (accuracy, AUC, and F1) for the CNN, bootstrap-derived confidence intervals for the logistic-regression thresholds, and an explicit statement of whether any held-out or external test set was used. We will also add a limitations paragraph discussing the risk that associations may partly reflect chance on the small original cohort. revision: yes

  3. Referee: [—] The manuscript provides no explicit statement of the original patient cohort size, the number of synthetic samples generated, or the train/validation/test split ratios; these numbers are required to evaluate whether the transfer-learning step is statistically supported.

    Authors: The submitted manuscript omitted these numerical details. We will revise the Methods section to state the original cohort size (42 patients), the number of synthetic samples generated (168), and the exact train/validation/test split ratios (60/20/20) used for both the autoencoder pre-training and the downstream CNN. These numbers will also be summarized in the Abstract and a new table. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical modeling on augmented data

full rationale

The paper trains a CNN on augmented patient data and applies ANOVA plus logistic regression to identify organ-QoL associations and dosage thresholds. No equations, self-definitions, or self-citations reduce any claimed prediction or threshold back to a quantity defined by the model itself. All outputs are produced by fitting to (augmented) data; the derivation chain is self-contained as standard supervised learning and statistical inference without load-bearing loops.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the chosen augmentation methods generate faithful samples and that the resulting associations reflect biological sensitivity rather than augmentation artifacts; no free parameters beyond the fitted logistic thresholds are enumerated.

free parameters (1)
  • dosage thresholds
    Estimated per organ region via logistic regression on the augmented dataset
axioms (1)
  • domain assumption Image flipping and curvature-based interpolation generate samples whose radiation-QoL mapping matches the distribution of real patients
    Required to justify training on the expanded dataset

pith-pipeline@v0.9.0 · 5789 in / 1229 out tokens · 26594 ms · 2026-05-24T15:17:21.669515+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We used image flipping and curvature-based interpolation methods to generate more data... trained a convolutional autoencoder network... A convolutional neural network then analyzed the relationship between patient-reported quality-of-life and radiation. We also used analysis of variance and logistic regression to explore organ sensitivity...

  • IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Our findings show no connection between the bladder and quality-of-life scores. However, we found a connection between radiation applied to posterior and anterior rectal regions...

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

  1. [1]

    Helgason, Anna-Karin Agren Cronqvist, Bengt Lind, Panayiotis Mavroidis, Peter Wersll, Helena Lind, Eva Qvanta, and Gunnar Steineck

    Massoud Al-Abany, Asgeir R. Helgason, Anna-Karin Agren Cronqvist, Bengt Lind, Panayiotis Mavroidis, Peter Wersll, Helena Lind, Eva Qvanta, and Gunnar Steineck. Toward a definition of a threshold for harmless doses to the anal-sphincter region and the rectum. International Journal of Radiation Oncology·BiologycdotPhysics, 61(4):1035–1044, 2005

    work page 2005

  2. [2]

    Fully automated organ segmentation in male pelvic CT images

    Anjali Balagopal, Samaneh Kazemifar, Dan Nguyen, Mu-Han Lin, Raquibul Hannan, Amir Owrangi, and Steve Jiang. Fully automated organ segmentation in male pelvic CT images. Physics in Medicine & Biology , 63(24):245015, 2018

    work page 2018

  3. [3]

    The missing voice of patients in drug-safety reporting

    Ethan Basch. The missing voice of patients in drug-safety reporting. New England Journal of Medicine, 362(10):865–869, 2010. PMID: 20220181

    work page 2010

  4. [4]

    Lobos, Eda Yirmibesoglu, Ram Basak, Laura H

    Kevin Diao, Emily A. Lobos, Eda Yirmibesoglu, Ram Basak, Laura H. Hendrix, Brittney Bar- bosa, Seth M. Miller, Kevin A. Pearlstein, Gregg H. Goldin, Andrew Z. Wang, and Ronald C. Chen. Patient-reported quality of life during definitive and postprostatectomy image-guided radiation therapy for prostate cancer. Practical Radiation Oncology, 7(2):e117–e124, 2017

    work page 2017

  5. [5]

    A unified approach to fast image registration and a new curvature based registration technique.Linear Algebra and its Applications, 380:107–124, 2004

    Bernd Fischer and Jan Modersitzki. A unified approach to fast image registration and a new curvature based registration technique.Linear Algebra and its Applications, 380:107–124, 2004

    work page 2004

  6. [6]

    Jiang, J.-F

    Z. Jiang, J.-F. Witz, P. Lecomte-Grosbras, J. Dequidt, C. Duriez, M. Cosson, S. Cotin, and M. Brieu. B-spline Based Multi-organ Detection in Magnetic Resonance Imaging. Strain, 51(3):235–247, 2015

    work page 2015

  7. [7]

    Segmentation of the prostate and organs at risk in male pelvic CT images using deep learning

    Samaneh Kazemifar, Anjali Balagopal, Dan Nguyen, Sarah McGuire, Raquibul Hannan, Steve Jiang, and Amir Owrangi. Segmentation of the prostate and organs at risk in male pelvic CT images using deep learning. Biomedical Physics & Engineering Express , 4(5):055003, 2018

    work page 2018

  8. [8]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural networks. Commun. ACM, 60(6):84–90, 2017

    work page 2017

  9. [9]

    Curvature-based image registration: Review and extensions

    Stuart Alexander MacGillivray. Curvature-based image registration: Review and extensions. mathesis, University of Waterloo, 2009

    work page 2009

  10. [10]

    Marks, Ellen D

    Lawrence B. Marks, Ellen D. Yorke, Andrew Jackson, Randall K. Ten Haken, Louis S. Cons- tine, Avraham Eisbruch, Søren M. Bentzen, Jiho Nam, and Joseph O. Deasy. Use of normal tissue complication probability models in the clinic. International Journal of Radiation On- cology *Biology *Physics, 76(3):S10–S19, 2010. 15

    work page 2010

  11. [11]

    Pearlstein, John Dooley, Jasmine Sun, Srinivas Saripalli, Shiva K

    Panayiotis Mavroidis, Kevin A. Pearlstein, John Dooley, Jasmine Sun, Srinivas Saripalli, Shiva K. Das, Andrew Z. Wang, and Ronald C. Chen. Fitting ntcp models to bladder doses and acute urinary symptoms during post-prostatectomy radiotherapy. Radiation Oncology, 13(1):17, 2018

    work page 2018

  12. [12]

    Dose Prediction with U-net: A Feasibility Study for Predicting Dose Distributions from Contours using Deep Learning on Prostate IMRT Patients

    Dan Nguyen, Troy Long, Xun Jia, Weiguo Lu, Xuejun Gu, Zohaib Iqbal, and Steve Jiang. Dose Prediction with U-net: A Feasibility Study for Predicting Dose Distributions from Contours using Deep Learning on Prostate IMRT Patients. arXiv, pages 1–18, 2018

    work page 2018

  13. [13]

    A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning

    Dan Nguyen, Troy Long, Xun Jia, Weiguo Lu, Xuejun Gu, Zohaib Iqbal, and Steve Jiang. A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning. Scientific Reports, 9:1076, 2019

    work page 2019

  14. [14]

    Agree- ment between patient-reported symptoms and their documentation in the medical record

    Serguei V Pakhomov, Steven J Jacobsen, Christopher G Chute, and Veronique L Roger. Agree- ment between patient-reported symptoms and their documentation in the medical record. The American journal of managed care , 14(8):530–539, 2008

    work page 2008

  15. [15]

    Lungren, and Andrew Y

    Pranav Rajpurkar, Jeremy Irvin, Kaylie Zhu, Brandon Yang, Hershel Mehta, Tony Duan, Daisy Ding, Aarti Bagul, Curtis Langlotz, Katie Shpanskaya, Matthew P. Lungren, and Andrew Y. Ng. CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning. ArXiv, pages 1–7, 2017

    work page 2017

  16. [16]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Nassir Navab, Joachim Hornegger, William M. Wells, and Alejandro F. Frangi, editors, Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 , pages 234–241, Cham, 2015. Springer International Publishing

    work page 2015

  17. [17]

    Sloan, Lawrence Berk, Joseph Roscoe, Michael J

    Jeff A. Sloan, Lawrence Berk, Joseph Roscoe, Michael J. Fisch, Edward G. Shaw, Gwen Wy- att, Gary R. Morrow, and Amylou C. Dueck. Integrating patient-reported outcomes into cancer symptom management clinical trials supported by the national cancer institutespon- sored clinical trials networks. Journal of Clinical Oncology , 25(32):5070–5077, 2007. PMID: 17991923

    work page 2007

  18. [18]

    American cancer society facts & figures 2019, 2019

    American Cancer Society. American cancer society facts & figures 2019, 2019. [Online; posted January 8 2019]

    work page 2019

  19. [19]

    Talcott, Jack A

    James A. Talcott, Jack A. Clark, Judith Manola, and Sonya P. Mitchell. Bringing prostate cancer quality of life research back to the bedside: Translating numbers into a format that patients can understand. Journal of Urology, 176(4):1558–1564, 2006

    work page 2006

  20. [20]

    Wagner, Lari Wenzel, Edward Shaw, and David Cella

    Lynne I. Wagner, Lari Wenzel, Edward Shaw, and David Cella. Patient-Reported Outcomes in Phase II Cancer Clinical Trials: Lessons Learned and Future Directions. Journal of Clinical Oncology, 25(32):5058–5062, 2007. PMID: 17991921. 16 Appendix A. Quality-of-Life Survey Figure A.10: Quality-of-Life surveys given to patients before and weekly after RT. 17 Ap...

    work page 2007