Pith. sign in

REVIEW 4 major objections 5 minor 53 references

SurLonFormer uses a Transformer over a patient's MRI history to predict time to Alzheimer's diagnosis and to locate the brain regions driving the risk.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

A Transformer combining vision and sequence encoders with a Cox survival head for dynamic survival prediction from longitudinal MRI, evaluated on Alzheimer's disease data.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Sensible transformer architecture for dynamic survival prediction from longitudinal images, but the biomarker-identification claim rests on a simulation sanity check and the real-data numbers are not visible in the preprint. the 4 major comments →

arxiv 2508.09328 v1 pith:LB2HHIXK submitted 2025-08-12 eess.IV cs.CVstat.APstat.OT

Dynamic Survival Prediction using Longitudinal Images based on Transformer

Bingfan Liu , Haolun Shi , Jiguo Cao This is my paper
classification eess.IV cs.CVstat.APstat.OT
keywords survival analysislongitudinal medical imagingTransformerCox proportional hazardsdynamic predictionAlzheimer's diseasecensored dataocclusion sensitivity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper tries to establish that a Transformer-based network, SurLonFormer, can use a patient's sequence of MRI scans plus clinical measurements to predict time to Alzheimer's diagnosis better than existing joint-model and CNN-LSTM baselines, while also pointing to the brain regions driving each prediction. It matters because survival analysis in medical imaging typically uses a single scan, ignores censoring, or cannot be updated as new scans arrive; SurLonFormer is designed to incorporate censored observations, capture correlations across visits, and give dynamic risk scores that can be recomputed at each new scan. In simulations and a real longitudinal Alzheimer's cohort, the paper reports higher C-index and AUC than baselines, and its occlusion-sensitivity maps align with known risky regions in synthetic data.

Core claim

SurLonFormer claims that survival risk from longitudinal imaging can be modeled end-to-end by a vision encoder that extracts spatial features from each scan, a sequence encoder that aggregates these features over visit times, and a Cox-based survival encoder that turns the combined representation into a risk score. The model is trained with a Cox partial-likelihood loss that uses censored patients, and at any landmark time it outputs a dynamic risk score based only on images observed up to that time. The paper reports that this architecture outperforms FPCA-Cox, LoFPCA-Cox, and CNN-LSTM in both synthetic experiments and the Alzheimer's cohort, and that occlusion sensitivity analysis on simul

What carries the argument

The central object is the three-stage architecture: (1) a Vision Encoder, an image feature extractor; (2) a Sequence Encoder, a transformer that models temporal dependence among features extracted at multiple visits; (3) a Survival Encoder, a Cox proportional hazards head that outputs a risk score. The claim that temporal correlations are captured rests on the sequence encoder's attention over visit times, and the claim that censoring is handled rests on the Cox partial-likelihood training objective.

Load-bearing premise

The interpretability claim assumes that the model's ability to find known risky regions in simulated images transfers to real MRI scans, where the true disease-related regions are not known.

What would settle it

If, on real Alzheimer's MRI scans, the occlusion-sensitivity maps highlight regions that trained radiologists would not associate with Alzheimer's atrophy, or if removing highlighted regions from the input changes the risk score no more than removing random regions, the interpretability claim would be refuted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Dynamic risk scores can be recomputed at each clinical visit, so a patient's predicted trajectory updates as new scans arrive.
  • Censored patients contribute to training through the Cox partial likelihood, using information up to their last follow-up.
  • The architecture scales to multiple images per patient, unlike FPCA-based joint models that have trouble with high-dimensional image data.
  • Occlusion sensitivity maps provide a biomarker-localization tool that could highlight regions driving a predicted transition to Alzheimer's disease.

Where Pith is reading between the lines

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

  • The occlusion-sensitivity validation is done on simulated images where the risky regions are known; a direct test on real scans would require comparing highlighted regions to expert-defined atrophy maps, which the paper does not report.
  • Because the sequence encoder is a transformer, the approach could be applied to other longitudinal image modalities such as CT or fundus photos without changing the architecture.
  • The reported improvement over CNN-LSTM might come more from handling censoring and temporal attention than from the vision encoder; ablations separating those contributions would clarify the mechanism.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes SurLonFormer, a Transformer-based architecture for dynamic survival prediction from longitudinal medical images combined with structured clinical data. The model consists of a Vision Encoder for spatial features, a Sequence Encoder for temporal aggregation, and a Survival Encoder built on the Cox proportional hazards model. The authors claim that SurLonFormer makes better use of censored data, handles scalability, and provides interpretability through occlusion sensitivity analysis. Evidence is presented via a simulation study, where synthetic images contain disease-related features in known 8x8 diagonal blocks, and a real-data application to ADNI MRI scans for predicting time to AD diagnosis. The abstract states that extensive simulations and the ADNI application demonstrate superior predictive performance and successful identification of disease-related imaging biomarkers.

Significance. If the central claims are supported, the paper would make a useful contribution to survival analysis with longitudinal imaging: the architecture jointly models spatial, temporal, and survival information, and the dynamic prediction framework is practically relevant for diseases like Alzheimer's. The simulation study is a sensible sanity-check design, and the inclusion of an interpretability mechanism is a positive feature. However, the evidentiary value of the current manuscript is limited because the real-data results are not visible in the text, the interpretability claim rests entirely on synthetic occlusion maps, and the simulation evaluation is qualitative rather than quantitative. The paper therefore holds promise but requires substantial strengthening before the claims can be accepted.

major comments (4)
  1. [Section 4 (Real Data Analysis)] The manuscript states that SurLonFormer achieves 'superior predictive performance' on ADNI, but the visible text contains no quantitative results: no C-index, AUC, Brier score, standard errors, confidence intervals, or fold-wise statistics are reported for any method. The comparison with FPCA-Cox, LoFPCA-Cox, and CNN-LSTM is mentioned only in generic terms (e.g., 'lower AUC and C-index values'), with no numbers. This is a load-bearing omission because the central claim of predictive superiority cannot be assessed. Please include a complete results table for all landmark times and time increments, with means and standard deviations across cross-validation folds, and preferably statistical tests or overlap intervals.
  2. [Section 3 (Simulation) and Figure 5] The interpretability claim—that SurLonFormer 'identifies disease-related imaging biomarkers'—is supported only by occlusion sensitivity maps on simulated images whose risk regions are defined by the authors' own data-generating process in Equation (4). The text explicitly says the highlighted regions 'align with the data-generating process,' which is a model-sanity check, not evidence that the model localizes actual disease-related structures in real MRI scans. In the ADNI analysis, no validation of the highlighted regions is reported: no overlap with anatomical ROIs, no correlation with FreeSurfer-based atrophy measures, and no comparison with established AD biomarkers. The claim should be either removed from the abstract or supported by real-data validation of the attention/occlusion maps.
  3. [Section 3 (Simulation)] The simulation study is described only qualitatively ('demonstrate superior predictive performance,' 'align with the data-generating process'). There is no quantitative summary (e.g., C-index/AUC/Brier values with Monte Carlo standard errors across repeated simulations) and no table recording performance across the six landmark/time-increment scenarios. Since the simulation is the main controlled validation of the method, the absence of numerical results makes it impossible to judge whether differences are meaningful or within noise. Please provide a simulation results table and specify the number of simulated datasets and patients.
  4. [Section 4 (Real Data Analysis) and Model Description] The paper reports that CNN-LSTM exhibits overfitting but gives no corresponding diagnostics for SurLonFormer. Given that dropout regularization is mentioned as a cause of the moderate off-diagonal sensitivities in Figure 5, it is important to report training/validation curves, the actual dropout rates used, and other hyperparameter settings (depth, heads, embedding dimension, learning rate). This is essential both for reproducibility and for assessing whether the reported performance is achieved in a stable regime.
minor comments (5)
  1. [Title page] Typo in affiliation: 'Acturial Science' should be 'Actuarial Science.' Also the preprint line says 'September 23, 2025' while the arXiv submission is dated August 12, 2025; please use a consistent, current date.
  2. [Notation] In several places the manuscript contains LaTeX placeholder tokens such as 'f �����g' for landmark times and time increments. These need to be filled in with the actual values before submission.
  3. [Figure 3 and 4] The architecture figures are described in the text but are not included in the visible manuscript text at the cited locations. Ensure all figures are embedded with clear captions and that labels (e.g., 'Sequence Encoder', 'Survival Encoder') are legible.
  4. [Section 1 (Introduction)] The reference list is not shown in the visible text. Please check that all cited works, especially [30, 51] for AD projections and [32, 33] for FreeSurfer, are complete and correctly formatted.
  5. [Section 4 (Real Data Analysis)] The description of the landmark prediction scheme is incomplete. Define clearly how the landmark time t*, the increment t, and the censoring at the landmark are handled in the risk set definition, and explain what 'dynamic' means in the evaluation protocol.

Circularity Check

0 steps flagged

No significant circularity: predictive performance is evaluated on held-out data against external baselines, and the synthetic occlusion-sensitivity analysis is a validation exercise, not a circular derivation.

full rationale

SurLonFormer's core predictive claim is tested by training on one portion of the data and evaluating on held-out validation sets, with comparisons to external baselines (FPCA-Cox, LoFPCA-Cox, CNN-LSTM) using standard metrics (AUC, C-index, Brier). There is no evidence that any parameter is fitted to the evaluation target or that a 'prediction' is reconstructed from the model's own inputs by construction. The occlusion sensitivity analysis in the simulation section uses a known data-generating process (Equation 4) as ground truth; the paper explicitly notes that the highlighted regions 'align with the data-generating process.' This is a legitimate sanity check on synthetic data, not a circular derivation, because the ground truth is defined independently of the model's learned representations. The real-data interpretability claim is indeed less strongly evidenced—no quantitative comparison with anatomical ROIs or independent AD biomarkers is reported—but that is an evidential limitation, not circularity. No load-bearing self-citations are identified in the provided text. The derivation chain is therefore self-contained with respect to circularity concerns.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

The central claim rests on the Cox modeling assumption, the comparability of the imaging sequences, non-informative censoring, and the transferability of simulation-based interpretability to real data. Model hyperparameters are undisclosed in the available text.

free parameters (1)
  • Model hyperparameters (architecture depth, attention heads, embedding dimension, dropout, learning rate)
    Not reported in the visible text; required to reproduce SurLonFormer's exact configuration.
axioms (4)
  • domain assumption Cox proportional hazards assumption applies to the neural-network-derived risk scores.
    The Survival Encoder uses the Cox model; the proportionality of hazards is assumed and not tested in the visible text.
  • domain assumption Affine-registered ADNI MRI scans are comparable across patients and visits.
    The preprocessing standardizes images to a common stereotaxic space, treating the image sequence as homogeneous input for the Transformer.
  • domain assumption Censoring is non-informative conditional on the observed features.
    Cox partial likelihood requires this; dropout and loss of follow-up are treated as conditionally independent of event time.
  • ad hoc to paper The simulation data-generating process (Equation 4) is representative of real AD progression.
    The model's interpretability is validated on synthetic images with known disease regions; the paper assumes this transfers to real ADNI images.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamic Survival Prediction using Longitudinal Images based on Transformer." pith.science (2026). https://pith.science/paper/LB2HHIXK

@misc{pith2026250809328,
  author       = {Pith},
  title        = {Pith review of: Dynamic Survival Prediction using Longitudinal Images based on Transformer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LB2HHIXK}},
  note         = {Machine review of arXiv:2508.09328}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Survival analysis utilizing multiple longitudinal medical images plays a pivotal role in the early detection and prognosis of diseases by providing insight beyond single-image evaluations. However, current methodologies often inadequately utilize censored data, overlook correlations among longitudinal images measured over multiple time points, and lack interpretability. We introduce SurLonFormer, a novel Transformer-based neural network that integrates longitudinal medical imaging with structured data for survival prediction. Our architecture comprises three key components: a Vision Encoder for extracting spatial features, a Sequence Encoder for aggregating temporal information, and a Survival Encoder based on the Cox proportional hazards model. This framework effectively incorporates censored data, addresses scalability issues, and enhances interpretability through occlusion sensitivity analysis and dynamic survival prediction. Extensive simulations and a real-world application in Alzheimer's disease analysis demonstrate that SurLonFormer achieves superior predictive performance and successfully identifies disease-related imaging biomarkers.

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

53 extracted references · 50 canonical work pages

  1. [1]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, Alexey, Beyer, Lucas, Kolesnikov, Alexander, Weissenborn, Dirk, Zhai, Xiaohua, Unterthiner, Thomas, Dehghani, Mostafa, Minderer, Matthias, Heigold, Georg, Gelly, Sylvain, Uszkoreit, Jakob, Houlsby, Neil. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. Proceedings of the International Conference on Learning Represen...

    work page 2021

  2. [2]

    A Time-dependent Discrimination Index for Survival Data

    Antolini, Laura, Boracchi, Patrizia, Biganzoli, Elia. A Time-dependent Discrimination Index for Survival Data. Statistics in Medicine, 24(24):3927--3944, 2005

    work page 2005

  3. [3]

    Estimation Methods for Time-dependent AUC Models with Survival data

    Hung, Hung, Chiang, Chin-Tsang. Estimation Methods for Time-dependent AUC Models with Survival data. Canadian Journal of Statistics, 38(1):8--26, 2010

    work page 2010

  4. [4]

    Quantifying and Comparing Dynamic Predictive Accuracy of Joint Models for Longitudinal Marker and Time-to-event in Presence of Censoring and Competing Risks

    Blanche, Paul, Proust-Lima, C \'e cile, Loubere, Lucie, Berr, Claudine, Dartigues, Jean-Fran c ois, Jacqmin-Gadda, H \'e l \`e ne. Quantifying and Comparing Dynamic Predictive Accuracy of Joint Models for Longitudinal Marker and Time-to-event in Presence of Censoring and Competing Risks. Biometrics, 71(1):102--113, 2015

    work page 2015

  5. [5]

    Attention Is All You Need

    Vaswani, Ashish, Shazeer, Noam, Parmar, Niki, Uszkoreit, Jakob, Jones, Llion, Gomez, Aidan, Kaiser, Lukasz, Polosukhin, Illia. Attention Is All You Need. Advances in Neural Information Processing Systems, 2017

    work page 2017

  6. [6]

    Review of Current Advances in Survival Analysis and Frailty Models

    Govindarajulu, Usha S, D'Agostino Sr, Ralph B. Review of Current Advances in Survival Analysis and Frailty Models. Wiley Interdisciplinary Reviews: Computational Statistics, 12(6):e1504, 2020

    work page 2020

  7. [7]

    Deep Survival Analysis

    Ranganath, Rajesh, Perotte, Adler, Elhadad, No \'e mie, Blei, David. Deep Survival Analysis. In: Machine Learning for Healthcare Conference, pp. 101--114, 2016

    work page 2016

  8. [8]

    Stochastic Calculus as a Tool in Survival Analysis: A Review

    McKeague, Ian W, Utikal, Klaus J. Stochastic Calculus as a Tool in Survival Analysis: A Review. Applied Mathematics and Computation, 39(3):155s--181s, 1990

    work page 1990

  9. [9]

    Deep Learning for Survival Analysis: A Review

    Wiegrebe, Simon, Kopper, Philipp, Sonabend, Raphael, Bischl, Bernd, Bender, Andreas. Deep Learning for Survival Analysis: A Review. Artificial Intelligence Review, 57(3):65, 2024

    work page 2024

  10. [10]

    Semiparametric Bayesian Analysis of Survival Data

    Sinha, Debajyoti, Dey, Dipak K. Semiparametric Bayesian Analysis of Survival Data. Journal of the American Statistical Association, 92(439):1195--1212, 1997

    work page 1997

  11. [11]

    Machine Learning for Survival Analysis: A Survey

    Wang, Ping, Li, Yan, Reddy, Chandan K. Machine Learning for Survival Analysis: A Survey. ACM Computing Surveys (CSUR), 51(6):1--36, 2019

    work page 2019

  12. [12]

    Survival Analysis: A Survey

    Chung, Ching-Fan, Schmidt, Peter, Witte, Ana D. Survival Analysis: A Survey. Journal of Quantitative Criminology, 7:59--98, 1991

    work page 1991

  13. [13]

    Survival analysis in public health research

    Lee, Elisa T, Go, Oscar T. Survival analysis in public health research. Annual Review of Public Health, 18(1):105--134, 1997

    work page 1997

  14. [14]

    Regression Models and Life-tables

    Cox, David R. Regression Models and Life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 34(2):187--202, 1972

    work page 1972

  15. [15]

    Multivariate Survival Analysis using Cox's Regression Model

    Christensen, Erik. Multivariate Survival Analysis using Cox's Regression Model. Hepatology, 7(6):1346--1358, 1987

    work page 1987

  16. [16]

    A General Framework for Random Effects Survival Analysis in the Cox Proportional Hazards Setting

    Sargent, Daniel J. A General Framework for Random Effects Survival Analysis in the Cox Proportional Hazards Setting. Biometrics, pp. 1486--1497, 1998

    work page 1998

  17. [17]

    Analysis of Survival Data under the Proportional Hazards Model

    Breslow, Norman E. Analysis of Survival Data under the Proportional Hazards Model. International Statistical Review/Revue Internationale de Statistique, pp. 45--57, 1975

    work page 1975

  18. [18]

    Accelerated Failure Time Models: A Review

    Dupuy, Jean-Fran c ois. Accelerated Failure Time Models: A Review. International Journal of Performability Engineering, 10(1):23, 2014

    work page 2014

  19. [19]

    Comparing Proportional Hazards and Accelerated Failure Time Models for Survival Analysis

    Orbe, Jesus, Ferreira, Eva, N \'u \ n ez-Ant \'o n, Vicente. Comparing Proportional Hazards and Accelerated Failure Time Models for Survival Analysis. Statistics in Medicine, 21(22):3493--3510, 2002

    work page 2002

  20. [20]

    Survival Analysis under the Cox Proportional Hazards Model with Pooled Covariates

    Saha-Chaudhuri, Paramita, Juwara, Lamin. Survival Analysis under the Cox Proportional Hazards Model with Pooled Covariates. Statistics in Medicine, 40(4):998--1020, 2021

    work page 2021

  21. [21]

    Joint Modeling of Survival and Longitudinal Data: Likelihood Approach Revisited

    Hsieh, Fushing, Tseng, Yi-Kuan, Wang, Jane-Ling. Joint Modeling of Survival and Longitudinal Data: Likelihood Approach Revisited. Biometrics, 62(4):1037--1043, 2006

    work page 2006

  22. [22]

    Basic Concepts and Methods for Joint Models of Longitudinal and Survival Data

    Ibrahim, Joseph G, Chu, Haitao, Chen, Liddy M. Basic Concepts and Methods for Joint Models of Longitudinal and Survival Data. Journal of Clinical Oncology, 28(16):2796--2801, 2010

    work page 2010

  23. [23]

    Joint Modeling of Longitudinal and Survival Data

    Wang, Jane-Ling, Zhong, Qixian. Joint Modeling of Longitudinal and Survival Data. Annual Review of Statistics and Its Application, 12, 2024

    work page 2024

  24. [24]

    Deep Survival Analysis with Longitudinal X-rays for COVID-19

    Shu, Michelle, Bowen, Richard Strong, Herrmann, Charles, Qi, Gengmo, Santacatterina, Michele, Zabih, Ramin. Deep Survival Analysis with Longitudinal X-rays for COVID-19. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 4046--4055, 2021

    work page 2021

  25. [25]

    Deep Recurrent Survival Analysis

    Ren, Kan, Qin, Jiarui, Zheng, Lei, Yang, Zhengyu, Zhang, Weinan, Qiu, Lin, Yu, Yong. Deep Recurrent Survival Analysis. In: Proceedings of the AAAI Conference on Artificial Intelligence, 33:4798--4805, 2019

    work page 2019

  26. [26]

    Dynamic-deephit: A Deep Learning Approach for Dynamic Survival Analysis with Competing Risks Based on Longitudinal Data

    Lee, Changhee, Yoon, Jinsung, Van Der Schaar, Mihaela. Dynamic-deephit: A Deep Learning Approach for Dynamic Survival Analysis with Competing Risks Based on Longitudinal Data. IEEE Transactions on Biomedical Engineering, 67(1):122--133, 2019

    work page 2019

  27. [27]

    Time-to-event Prediction with Neural Networks and Cox Regression

    Kvamme, H vard, Borgan, rnulf, Scheel, Ida. Time-to-event Prediction with Neural Networks and Cox Regression. Journal of Machine Learning Research, 20(129):1--30, 2019

    work page 2019

  28. [28]

    Deep Learning for the Dynamic Prediction of Multivariate Longitudinal and Survival Data

    Lin, Jeffrey, Luo, Sheng. Deep Learning for the Dynamic Prediction of Multivariate Longitudinal and Survival Data. Statistics in Medicine, 41(15):2894--2907, 2022

    work page 2022

  29. [29]

    Multi-task Deep Learning-based Survival Analysis on the Prognosis of Late AMD using the Longitudinal Data in AREDS

    Ghahramani, Gregory, Brendel, Matthew, Lin, Mingquan, Chen, Qingyu, Keenan, Tiarnan, Chen, Kun, Chew, Emily, Lu, Zhiyong, Peng, Yifan, Wang, Fei. Multi-task Deep Learning-based Survival Analysis on the Prognosis of Late AMD using the Longitudinal Data in AREDS. In: AMIA Annual Symposium Proceedings, 2021:506, 2022

    work page 2021

  30. [30]

    Forecasting the Global Burden of Alzheimer’s Disease

    Brookmeyer, Ron, Johnson, Elizabeth, Ziegler-Graham, Kathryn, Arrighi, H Michael. Forecasting the Global Burden of Alzheimer’s Disease. Alzheimer's & Dementia, 3(3):186--191, 2007

    work page 2007

  31. [31]

    Whitwell, Jennifer, Ward, Chadwick, others

    Jack Jr, Clifford R, Bernstein, Matt A, Fox, Nick C, Thompson, Paul, Alexander, Gene, Harvey, Danielle, Borowski, Bret, Britson, Paula J, L. Whitwell, Jennifer, Ward, Chadwick, others. The Alzheimer's Disease Neuroimaging Initiative (ADNI): MRI Methods. Journal of Magnetic Resonance Imaging, 27(4):685--691, 2008

    work page 2008

  32. [32]

    Quantitative Assessment of Field Strength, Total Intracranial Volume, Sex, and Age Effects on the Goodness of Harmonization for Volumetric Analysis on the ADNI Database

    Ma, Da, Popuri, Karteek, Bhalla, Mahadev, Sangha, Oshin, Lu, Donghuan, Cao, Jiguo, Jacova, Claudia, Wang, Lei, Beg, Mirza Faisal, Alzheimer's Disease Neuroimaging Initiative. Quantitative Assessment of Field Strength, Total Intracranial Volume, Sex, and Age Effects on the Goodness of Harmonization for Volumetric Analysis on the ADNI Database. Human Brain ...

    work page 2019

  33. [33]

    Popuri, Karteek, Ma, Da, Wang, Lei, Beg, Mirza Faisal. Using Machine Learning to Quantify Structural MRI Neurodegeneration Patterns of Alzheimer's Disease into Dementia Score: Independent Validation on 8,834 Images from ADNI, AIBL, OASIS, and MIRIAD databases. Human Brain Mapping, 41(14):4127--4147, 2020

    work page 2020

  34. [34]

    Dynamic Survival Prediction using Sparse Longitudinal Images via Multi-dimensional Functional Principal Component Analysis

    Shi, Haolun, Jiang, Shu, Ma, Da, Beg, Mirza Faisal, Cao, Jiguo. Dynamic Survival Prediction using Sparse Longitudinal Images via Multi-dimensional Functional Principal Component Analysis. Journal of Computational and Graphical Statistics, 33(4):1--18, 2024

    work page 2024

  35. [35]

    Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling

    Holste, Gregory, Lin, Mingquan, Zhou, Ruiwen, Wang, Fei, Liu, Lei, Yan, Qi, Van Tassel, Sarah H, Kovacs, Kyle, Chew, Emily Y, Lu, Zhiyong, others. Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling. NPJ Digital Medicine, 7(1):216, 2024

    work page 2024

  36. [36]

    Counting processes and survival analysis

    Fleming, Thomas R, Harrington, David P. Counting processes and survival analysis. John Wiley & Sons, 2013

    work page 2013

  37. [37]

    Joint models for longitudinal and time-to-event data: With applications in R

    Rizopoulos, Dimitris. Joint models for longitudinal and time-to-event data: With applications in R. CRC Press, 2012

    work page 2012

  38. [38]

    Improved dynamic predictions from joint models of longitudinal and survival data with time-varying effects using P-splines

    Andrinopoulou, Eleni-Rosalina, Eilers, Paul HC, Takkenberg, Johanna JM, Rizopoulos, Dimitris. Improved dynamic predictions from joint models of longitudinal and survival data with time-varying effects using P-splines. Biometrics, 74(2):685--693, 2018

    work page 2018

  39. [39]

    Functional linear regression analysis for longitudinal data

    Yao, Fang, M \"u ller, Hans-Georg, Wang, Jane-Ling. Functional linear regression analysis for longitudinal data. The Annals of Statistics, 33(6), 2005

    work page 2005

  40. [40]

    A survey on deep learning in medical image analysis

    Litjens, Geert, Kooi, Thijs, Bejnordi, Babak Ehteshami, Setio, Arnaud Arindra Adiyoso, Ciompi, Francesco, Ghafoorian, Mohsen, Van Der Laak, Jeroen Awm, Van Ginneken, Bram, S \'a nchez, Clara I. A survey on deep learning in medical image analysis. Medical Image Analysis, 42:60--88, 2017

    work page 2017

  41. [41]

    DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network

    Katzman, Jared L, Shaham, Uri, Cloninger, Alexander, Bates, Jonathan, Jiang, Tingting, Kluger, Yuval. DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Medical Research Methodology, 18:1--12, 2018

    work page 2018

  42. [42]

    Deep learning for patient-specific kidney graft survival analysis

    Luck, Margaux, Sylvain, Tristan, Cardinal, H \'e lo \" se, Lodi, Andrea, Bengio, Yoshua. Deep learning for patient-specific kidney graft survival analysis. ArXiv Preprint ArXiv:1705.10245, 2017

    Pith/arXiv arXiv 2017

  43. [43]

    Feed forward neural networks for the analysis of censored survival data: a partial logistic regression approach

    Biganzoli, Elia, Boracchi, Patrizia, Mariani, Luigi, Marubini, Ettore. Feed forward neural networks for the analysis of censored survival data: a partial logistic regression approach. Statistics in Medicine, 17(10):1169--1186, 1998

    work page 1998

  44. [44]

    Learning transferable visual models from natural language supervision

    Radford, Alec, Kim, Jong Wook, Hallacy, Chris, Ramesh, Aditya, Goh, Gabriel, Agarwal, Sandhini, Sastry, Girish, Askell, Amanda, Mishkin, Pamela, Clark, Jack, others. Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning, pp. 8748--8763, 2021

    work page 2021

  45. [45]

    Rnn-surv: A deep recurrent model for survival analysis

    Giunchiglia, Eleonora, Nemchenko, Anton, van der Schaar, Mihaela. Rnn-surv: A deep recurrent model for survival analysis. In: Artificial Neural Networks and Machine Learning--ICANN 2018: 27th International Conference on Artificial Neural Networks, Rhodes, Greece, October 4-7, 2018, Proceedings, Part III 27, pp. 23--32, 2018

    work page 2018

  46. [46]

    Spatial-temporal transformer networks for traffic flow forecasting

    Xu, Mingxing, Dai, Wenrui, Liu, Chunmiao, Gao, Xing, Lin, Weiyao, Qi, Guo-Jun, Xiong, Hongkai. Spatial-temporal transformer networks for traffic flow forecasting. ArXiv Preprint ArXiv:2001.02908, 2020

    Pith/arXiv arXiv 2001

  47. [47]

    EEG temporal--spatial transformer for person identification

    Du, Yang, Xu, Yongling, Wang, Xiaoan, Liu, Li, Ma, Pengcheng. EEG temporal--spatial transformer for person identification. Scientific Reports, 12(1):14378, 2022

    work page 2022

  48. [48]

    Transformer-based spatial-temporal feature learning for EEG decoding

    Song, Yonghao, Jia, Xueyu, Yang, Lie, Xie, Longhan. Transformer-based spatial-temporal feature learning for EEG decoding. ArXiv Preprint ArXiv:2106.11170, 2021

    Pith/arXiv arXiv 2021

  49. [49]

    Visualizing and understanding convolutional networks

    Zeiler, Matthew D, Fergus, Rob. Visualizing and understanding convolutional networks. In: Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part I 13, pp. 818--833, 2014

    work page 2014

  50. [50]

    Deephit: A deep learning approach to survival analysis with competing risks

    Lee, Changhee, Zame, William, Yoon, Jinsung, Van Der Schaar, Mihaela. Deephit: A deep learning approach to survival analysis with competing risks. In: Proceedings of the AAAI conference on artificial intelligence, 32(1), 2018

    work page 2018

  51. [51]

    World Alzheimer Report 2015

    Prince, Martin, Wimo, Anders, Guerchet, Ma \"e lenn, Ali, Gemma-Claire, Wu, Yu-Tzu, Prina, Matthew. World Alzheimer Report 2015. The Global Impact of Dementia: An analysis of prevalence, incidence, cost and trends. Alzheimer's Disease International, 2015

    work page 2015

  52. [52]

    Joint modeling of longitudinal and time-to-event data: an overview

    Tsiatis, Anastasios A, Davidian, Marie. Joint modeling of longitudinal and time-to-event data: an overview. Statistica Sinica, pp. 809--834, 2004

    work page 2004

  53. [53]

    Joint models for dynamic prediction in localised prostate cancer: a literature review

    Parr, Harry, Hall, Emma, Porta, Nuria. Joint models for dynamic prediction in localised prostate cancer: a literature review. BMC Medical Research Methodology, 22(1):245, 2022

    work page 2022

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.