Pith. sign in

REVIEW 3 major objections 5 minor 49 references

Regulating radiology AI medical devices that evolve in their lifecycle

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read New US and EU regulatory pathways allow radiology AI devices to be updated after deployment without re-approval, provided manufacturers document planned changes and monitoring at initial clearance.

desk verdict Useful synthesis of the FDA PCCP and EU AI Act for radiology AI, but it overstates the EU side: the AI Act does not create a PCCP-like pre-approved change plan, and that legal error undercuts the comparative framing. read the letter →

arxiv 2412.20498 v3 pith:3I4B5QRY submitted 2024-12-29 cs.CY

Camila González , Moritz Fuchs , Daniel Pinto dos Santos , Philipp Matthies , Manuel Trenz , Maximilian Grüning , Akshay Chaudhari , David B. Larson
show 4 more authors
Ahmed Othman Moon Kim Felix Nensa Anirban Mukhopadhyay
This is my paper · ORCID
classification cs.CY
keywords AImedicaldevicesregulationcontinuallearningPredeterminedChangeControlPlanEUActradiologyperformancemonitoringSoftwareasaDevice
verification ladder T0 review T1 audit T2 compute T3 formal

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 argues that two 2024 regulatory changes—the European Union's Artificial Intelligence Act, in force since August, and the U.S. FDA's final guidance on a Predetermined Change Control Plan (PCCP), issued in December—create a legal pathway for radiology AI devices to be updated after deployment without going through full re-approval. The condition is that the manufacturer documented the planned changes, the retraining process, and the monitoring scheme at initial clearance, and that each update stays within that approved protocol. The authors further claim that five technical and organizational building blocks suffice to make such "lifelong" systems safe: structured reporting, continual learning methods, real-world performance monitoring, uncertainty quantification, and active radiologist involvement. A sympathetic reader should care because locked models are known to degrade as scanners, populations, and disease patterns drift, and the previous regime of expensive re-approval made timely updates rare.

What carries the argument

The central mechanism is the change control plan submitted at the time of initial clearance. It consists of change specifications (description of modifications, implementation protocol, impact assessment) plus a quality monitoring scheme, and it turns the device from a "locked" system into a "lifelong" one: approved updates can be deployed without re-approval, while deviations from the plan or changes in intended use still require traditional review. The paper then identifies five building blocks that carry the safety argument: structured reporting for label stability, continual learning for backward and forward transfer, predefined performance metrics and statistical tests, uncertainty quantification and out-of-distribution detection, and radiologist involvement as a monitoring and correction layer.

What would settle it

A prospective or retrospective study that measures the lag between the onset of a distribution shift in a deployed radiology AI and the moment its monitoring system raises an alarm, and compares that lag with the time in which model errors change clinical decisions, would settle the central safety premise. If alarms typically arrive after harm has occurred, the no-reapproval pathway does not deliver the safety it promises.

Watch

Extended reading notes

Core claim

The central claim is that a lifecycle regulatory protocol now exists on both sides of the Atlantic: instead of re-approving every model change, regulators accept a change control plan describing the modifications, protocol, and impact assessment, and then let manufacturers roll out updates that comply with it. The paper asserts that dynamic systems built on this pathway can be safe and effective if they combine structured reporting to keep labels consistent, continual learning techniques that avoid catastrophic forgetting, pre-defined standards and statistical tests for assessing performance over time, uncertainty quantification and out-of-distribution detection so the system knows when it cannot be trusted, and radiologists who curate data and act as an on-site quality assurance loop. The authors present this not as a purely theoretical possibility but as the direction regulators have already chosen, with the caveat that the guidance still leaves metrics, deterioration thresholds, and sub-population trade-offs undefined.

Load-bearing premise

The whole approach assumes that real-world monitoring can detect a performance drop early enough to act, but the paper concedes that by the time a deterioration is identified, the product must be taken out of routine use.

Editorial extensions

If this is right

  • Manufacturers can ship routine retrained models without re-approval, so the time between a model becoming obsolete and a fix reaching clinics can shrink from years to weeks.
  • Real-world performance monitoring stops being optional; every PCCP or AI Act submission will have to specify metrics, sub-populations, and statistical tests in advance.
  • Structured reporting becomes a practical prerequisite for model updating, because only standardized labels can be reused as training data without introducing annotation drift.
  • Continual learning methods such as rehearsal-based replay and pseudo-rehearsal become regulatory tools, not just research topics, since they let models improve on local data without full retraining.
  • Regulatory guidance gaps on thresholds and trade-offs mean that early adopters will effectively set the standards by discretion, until objective benchmarks appear.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The safety case hinges on monitoring lag: if performance degradation is detected only after it has caused harm, the no-reapproval pathway delivers no safety advantage over locked devices, a tension the paper itself notes but does not resolve.
  • Manufacturers may be incentivized to write broad change control plans to keep future flexibility, shifting the real gatekeeping burden to the initial review of what counts as "planned."
  • The same lifecycle logic should extend beyond radiology to other medical AI domains, but radiology's structured reporting culture makes it the likely first testbed, so evidence of monitoring lag elsewhere may come later.
  • A testable consequence is that devices with onboard uncertainty quantification and out-of-distribution detection should show shorter harmful-degradation windows than devices relying only on periodic external monitoring.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper argues that the 2024 FDA final guidance on Predetermined Change Control Plans (PCCP) and the EU Artificial Intelligence Act (AI Act) together establish a new lifecycle regulatory pathway for radiology AI software as a medical device: manufacturers can document planned changes at approval and roll them out, if they follow the protocol, without re-approval. It reviews the requirements these frameworks impose (real-world performance monitoring, privacy protection, bias mitigation, transparency, traceability), and it proposes five technical building blocks: structured reporting, continual learning, performance standards over time, uncertainty quantification and out-of-distribution detection, and active radiologist involvement. The paper concludes with a list of advantages and open challenges. It is a perspective/overview paper rather than an empirical study.

Significance. If its comparative legal premise were accurate, the paper would be a useful and timely synthesis connecting regulatory developments to concrete engineering requirements. Its strengths are the accessible description of the FDA PCCP, the clear Figure 2, and the balanced enumeration of benefits and risks in Figure 4. The paper also gathers a relevant set of technical references and identifies open questions (metrics, thresholds, trade-offs) that the community needs to address. However, because the central claim about the EU AI Act is not supported by the cited legal framework, the significance of the comparison is currently compromised.

major comments (3)
  1. [Section 2 (and Figure 2; Conclusions)] The paper's central claim that both the FDA PCCP and the EU AI Act allow manufacturers to document planned changes at approval and roll them out without re-approval is not supported for the EU. Regulation (EU) 2024/1689 (AI Act) defines 'substantial modification' in Article 3(23) and requires a new conformity assessment for any such modification under Article 43(4); it contains no mechanism analogous to the FDA's PCCP that grants advance approval to a class of future changes. The MDR (Regulation (EU) 2017/745) likewise treats significant changes as triggering renewed notified-body assessment. The 'commonalities' paragraph in Section 2 and the dual-jurisdiction workflow in Figure 2 therefore attribute a US-specific construct to EU law. The authors should either correct the EU description and limit the change-control-plan claim to the FDA, or present the EU as lacking an equivalent mechanism and argue for one, citing the exact articles.
  2. [Section 1 (constraint 4) and Section 2 ('Real-world performance monitoring')] The safety case for the lifelong pathway relies on continuous monitoring to detect performance degradation in time to act, but Section 1 states that 'by the time a performance deterioration has been identified, it is too late: The product must be taken out of the routine until an update has been conducted.' Section 2 then makes continuous monitoring the core safeguard of both regulatory models without resolving this tension. The paper does not explain how its proposed building blocks (uncertainty quantification, out-of-distribution detection, or radiologist involvement) provide early warning, nor does it discuss how monitoring thresholds would be set to allow intervention before harm. Without this, the claim that the new workflow 'would ensure the safety and effectiveness of the product' (Section 2) is unsupported.
  3. [Section 3 ('Standards for assessing performance over time')] The paper acknowledges that the regulations do not clarify what constitutes a performance increase or deterioration and then leaves threshold selection, metric weighting, and trade-off handling entirely to manufacturers. Given the paper's stated aim to outline 'the key building blocks necessary for successfully deploying dynamic systems,' this section provides a checklist rather than guidance; it does not address, for example, how to choose statistical tests with sufficient power for rare-event or sub-population monitoring, or how to act when metrics conflict. This gap is not fatal for a perspective piece, but it weakens the prescriptive claim that the identified building blocks are sufficient.
minor comments (5)
  1. [Section 2] The legal instruments are misdescribed: the MDR and the AI Act are directly applicable Regulations, not 'area-specific directives' or 'cross-sectional guidelines'; the term 'directive' has a specific meaning in EU law.
  2. [References] The reference for the AI Act (European Commission, 2021b) points to the final text in the Official Journal but the year '2021' is incorrect; the final Regulation was published in 2024. The stale proposal reference (2021a) should be removed or clearly marked as historical.
  3. [Figure 2] The figure shows 'APPROVAL' twice and uses the same visual style for initial clearance and the PCCP-based approval; the reader cannot immediately distinguish the two pathways, undermining the figure's purpose.
  4. [Section 3, 'Interoperability through structured reporting'] The statement that 'when all collected data follows the same format and degree of abstraction, new cases can be used to train the model without further preparation' is too strong, because radiology reports used for training still require verification against a reference standard; the paper should acknowledge that structured reporting alone does not produce labels.
  5. [Section 1] The claim that 'deep learning models... the performance of models on new data deteriorates until they are no longer reliable' is stated as a general law; the cited references demonstrate susceptibility to distribution shift, but the monotonic-deterioration narrative is an oversimplification and should be softened.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the paper's central claims are summaries of external regulatory documents and independent technical literature, not derivations from its own inputs.

full rationale

This paper is a review and position paper rather than a derivation. Its central claim, that the FDA PCCP and the EU AI Act create a lifecycle workflow permitting planned modifications to be rolled out without re-approval, is grounded in external legal sources (FDA 2023, 2024; European Commission 2021a,b) and is not obtained by fitting parameters, defining terms in terms of the conclusion, or renaming a known result. No quantity is fitted and then relabeled as a prediction, and no equation or protocol is defined in terms of its own target. The technical building blocks (structured reporting, continual learning, performance monitoring, uncertainty quantification, and radiologist involvement) are presented as prerequisites, supported by a mixture of external literature and the authors' prior works such as González et al. 2022a,b and Fuchs et al. 2022, 2023. Those self-citations support peripheral methodological illustrations rather than serving as load-bearing uniqueness arguments, so they do not make the regulatory analysis circular. The paper itself notes limitations, including Section 1 constraint 4 that by the time performance deterioration is identified it is too late, and Section 4's acknowledgement that critical gaps such as evaluation metrics and deterioration thresholds remain undefined; these are substantive weaknesses or legal-correctness concerns, not circular reasoning. The skeptic's objection that the EU AI Act may not actually provide a PCCP-like exemption is a challenge to the accuracy of the paper's reading of external law, not an instance of the paper deriving its conclusion from itself. Therefore, no specific circular step can be exhibited, and the appropriate finding is no significant circularity with a low score reflecting only minor non-load-bearing self-citation.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The paper's argument rests on several domain assumptions about model drift, data availability, label stability, monitoring effectiveness, and the content of the new regulatory guidance. No parameters are fitted and no new entities are introduced.

assumptions (6)
  • domain assumption Deep learning models are highly susceptible to even slight variations in image characteristics and deteriorate in performance over time.
    Invoked in Section 1 to motivate the need for lifecycle regulation; no systematic evidence is provided in this paper beyond general citations.
  • domain assumption Manufacturers continue to collect data beyond initial deployment, and this data is suitable and available for retraining.
    Assumed in Section 1; the paper itself notes constraints such as GDPR and HIPAA data deletion and computational cost that may make retraining impossible.
  • domain assumption Structured reporting with a fixed template keeps labeling distributions stable over time.
    Section 3 claims this as a prerequisite; the paper cites correlational evidence but no controlled validation that templates eliminate label shift.
  • domain assumption Real-world performance monitoring can detect degradation before patient harm.
    Section 2 makes this the central safeguard of PCCP and AI Act, yet Section 1 constraint 4 states the opposite timing problem.
  • domain assumption The FDA PCCP final guidance and EU AI Act impose the requirements described in Figure 1.
    The paper summarizes the regulations but does not quote or reproduce the legal text; readers must trust the authors' reading of FDA 2024 and European Commission 2021b.
  • domain assumption The notified-body backlog in the EU is a major cause of infrequent updates.
    Section 1 relies on a commercial blog, Open Regulatory 2023, for this load-bearing claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Regulating radiology AI medical devices that evolve in their lifecycle." pith.science (2026). https://pith.science/paper/3I4B5QRY

@misc{pith2026241220498,
  author       = {Pith},
  title        = {Pith review of: Regulating radiology AI medical devices that evolve in their lifecycle},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3I4B5QRY}},
  note         = {Machine review of arXiv:2412.20498}
}
read the original abstract

Over time, the distribution of medical image data drifts due to factors such as shifts in patient demographics, acquisition devices, and disease manifestations. While human radiologists can adjust their expertise to accommodate such variations, deep learning models cannot. In fact, such models are highly susceptible to even slight variations in image characteristics. Consequently, manufacturers must conduct regular updates to ensure that they remain safe and effective. Performing such updates in the United States and European Union required, until recently, obtaining re-approval. Given the time and financial burdens associated with these processes, updates were infrequent, and obsolete systems remained in operation for too long. During 2024, several regulatory developments promised to streamline the safe rollout of model updates: The European Artificial Intelligence Act came into effect last August, and the Food and Drug Administration (FDA) issued final marketing submission recommendations for a Predetermined Change Control Plan (PCCP) in December. We provide an overview of these developments and outline the key building blocks necessary for successfully deploying dynamic systems. At the heart of these regulations - and as prerequisites for manufacturers to conduct model updates without re-approval - are clear descriptions of data collection and re-training processes, coupled with robust real-world quality monitoring mechanisms.

Figures

Figures reproduced from arXiv: 2412.20498 by the authors.

Figure 1
Figure 1. From bottom to top: Requirements set by recent regulatory e [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Approval process for a deterministic locked AI system (top) vs. a lifelong learning system (bottom). In the second case, a description of the planned modifications and protocols outlining how they will be implemented and monitored are submitted as additional documentation during initial clearance. Modifications following these specifications, such as re-training the model with additional training data, do not requir… view at source ↗
Figure 3
Figure 3. Continuously learning systems adapt to new environments, obtaining better performance on local data distributions. By [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Advantages and disadvantages of locked vs. continual learning systems with respect to resource utilization and model [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 47 canonical work pages

  1. [1]

    , year 1996

    author Act, A. , year 1996 . title Health insurance portability and accountability act of 1996 . journal Public law volume 104 , pages 191

    work page 1996

  2. [2]

    , year 2022

    author Alvarado, R. , year 2022 . title Should we replace radiologists with deep learning? pigeons, error and trust in medical ai . journal Bioethics volume 36 , pages 121--133

    work page 2022

  3. [3]

    , author Gaw, N

    author Arun, N. , author Gaw, N. , author Singh, P. , author Chang, K. , author Aggarwal, M. , author Chen, B. , author Hoebel, K. , author Gupta, S. , author Patel, J. , author Gidwani, M. , et al., year 2021 . title Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging . journal Radiology: Artificial Intelligence...

    work page 2021

  4. [4]

    , author Baylor, E

    author Beede, E. , author Baylor, E. , author Hersch, F. , author Iurchenko, A. , author Wilcox, L. , author Ruamviboonsuk, P. , author Vardoulakis, L.M. , year 2020 . title A human-centered evaluation of a deep learning system deployed in clinics for the detection of diabetic retinopathy , in: booktitle Proceedings of the 2020 CHI conference on human fac...

    work page 2020

  5. [5]

    , author Kitamura, F.C

    author Colak, E. , author Kitamura, F.C. , author Hobbs, S.B. , author Wu, C.C. , author Lungren, M.P. , author Prevedello, L.M. , author Kalpathy-Cramer, J. , author Ball, R.L. , author Shih, G. , author Stein, A. , et al., year 2021 . title The rsna pulmonary embolism ct dataset . journal Radiology: Artificial Intelligence volume 3 , pages e200254

    work page 2021

  6. [6]

    author European Commission , year 2021 a. title Proposal for a regulation of the european parliament and the council laying down harmonised rules on artificial intelligence (artificial intelligence act) and amending certain union legislative acts . https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021PC0206&from=EN. note accessed: 2023-01-25

    work page 2021

  7. [7]

    title Regulation of the european parliament and the council laying down harmonised rules on artificial intelligence (artificial intelligence act)

    author European Commission , year 2021 b. title Regulation of the european parliament and the council laying down harmonised rules on artificial intelligence (artificial intelligence act) . https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ

    work page 2021

  8. [8]

    title Regulation (eu) 2017/745 of the european parliament and of the council of 5 april 2017 on medical devices

    author European Parliament and Council , year 2017 . title Regulation (eu) 2017/745 of the european parliament and of the council of 5 april 2017 on medical devices . howpublished Available at https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX\ note Official Journal of the European Union, L 117, 5 May 2017, pp. 1-175

    work page 2017

Show all 49 references
  1. [9]

    author FDA , year 2023 . title Marketing submission recommendations for a predetermined change control plan for artificial intelligence/machine learning (ai/ml)-enabled device software functions - draft guidance for industry and food and drug administration staff

    2023

  2. [10]

    title Marketing submission recommendations for a predetermined change control plan for artificial intelligence-enabled device software functions

    author FDA , year 2024 . title Marketing submission recommendations for a predetermined change control plan for artificial intelligence-enabled device software functions . howpublished https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submissio...

    2024

  3. [11]

    author FDA , et al., year 2019 . title Proposed regulatory framework for modifications to artificial intelligence/machine learning (ai/ml)-based software as a medical device (samd) https://www.fda.gov/files/medical\

    2019

  4. [12]

    title Artificial intelligence/machine learning (ai/ml)-based software as a medical device (samd) action plan

    author FDA , et al., year 2021 . title Artificial intelligence/machine learning (ai/ml)-based software as a medical device (samd) action plan . journal US Food Drug Admin., White Oak, MD, USA, Tech. Rep volume 145022 . https://www.fda.gov/media/145022/download

    2021

  5. [13]

    , author Gonzalez, C

    author Fuchs, M. , author Gonzalez, C. , author Frisch, Y. , author Hahn, P. , author Matthies, P. , author Gruening, M. , author Pinto dos Santos, D. , author Dratsch, T. , author Kim, M. , author Nensa, F. , author Trenz, Manuel; Mukhopadhyay, A. , year 2023 . title Closing ...

    2023 doi

  6. [14]

    , author Gonz \'a lez, C

    author Fuchs, M. , author Gonz \'a lez, C. , author Mukhopadhyay, A. , year 2022 . title Practical uncertainty quantification for brain tumor segmentation , in: booktitle International Conference on Medical Imaging with Deep Learning , organization PMLR . pp. pages 407--422

    2022

  7. [15]

    , author Duong, P.A.T

    author Ganeshan, D. , author Duong, P.A.T. , author Probyn, L. , author Lenchik, L. , author McArthur, T.A. , author Retrouvey, M. , author Ghobadi, E.H. , author Desouches, S.L. , author Pastel, D. , author Francis, I.R. , year 2018 . title Structured reporting in radiology ....

    2018

  8. [16]

    , author Jacobsen, J.H

    author Geirhos, R. , author Jacobsen, J.H. , author Michaelis, C. , author Zemel, R. , author Brendel, W. , author Bethge, M. , author Wichmann, F.A. , year 2020 . title Shortcut learning in deep neural networks . journal Nature Machine Intelligence volume 2 , pages 665--673

    2020

  9. [17]

    , author Gotkowski, K

    author Gonz \'a lez, C. , author Gotkowski, K. , author Fuchs, M. , author Bucher, A. , author Dadras, A. , author Fischbach, R. , author Kaltenborn, I.J. , author Mukhopadhyay, A. , year 2022 a. title Distance-based detection of out-of-distribution silent failures for covid-1...

    2022

  10. [18]

    , author Mukhopadhyay, A

    author Gonz \'a lez, C. , author Mukhopadhyay, A. , year 2021 . title Self-supervised out-of-distribution detection for cardiac cmr segmentation , in: booktitle International Conference on Medical Imaging with Deep Learning , organization PMLR . pp. pages 205--218

    2021

  11. [19]

    , author Ranem, A

    author Gonz \'a lez, C. , author Ranem, A. , author Othman, A. , author Mukhopadhyay, A. , year 2022 b. title Task-agnostic continual hippocampus segmentation for smooth population shifts , in: booktitle MICCAI Workshop on Domain Adaptation and Representation Transfer , organi...

    2022

  12. [20]

    , author Gonz \'a lez, C

    author Gotkowski, K. , author Gonz \'a lez, C. , author Kaltenborn, I. , author Fischbach, R. , author Bucher, A. , author Mukhopadhyay, A. , year 2022 . title i3deep: Efficient 3d interactive segmentation with the nnu-net , in: booktitle International Conference on Medical Im...

    2022

  13. [21]

    , author Wu, Z

    author Goyal, Y. , author Wu, Z. , author Ernst, J. , author Batra, D. , author Parikh, D. , author Lee, S. , year 2019 . title Counterfactual visual explanations , in: booktitle International Conference on Machine Learning , organization PMLR . pp. pages 2376--2384

    2019

  14. [22]

    , author Rao, D

    author Hadsell, R. , author Rao, D. , author Rusu, A.A. , author Pascanu, R. , year 2020 . title Embracing change: Continual learning in deep neural networks . journal Trends in cognitive sciences volume 24 , pages 1028--1040

    2020

  15. [23]

    , author Basart, S

    author Hendrycks, D. , author Basart, S. , author Mu, N. , author Kadavath, S. , author Wang, F. , author Dorundo, E. , author Desai, R. , author Zhu, T. , author Parajuli, S. , author Guo, M. , et al., year 2021 . title The many faces of robustness: A critical analysis of out...

    2021

  16. [24]

    title Software as a medical device (samd): Key definitions

    author IMDRF , year 2013 . title Software as a medical device (samd): Key definitions . https://www.imdrf.org/sites/default/files/docs/imdrf/final/technical/imdrf-tech-131209-samd-key-definitions-140901.pdf

    2013

  17. [25]

    , author Murray, N

    author Jetley, S. , author Murray, N. , author Vig, E. , year 2016 . title End-to-end saliency mapping via probability distribution prediction , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 5753--5761

    2016

  18. [26]

    , author Makowski, M.R

    author Kaissis, G.A. , author Makowski, M.R. , author R \"u ckert, D. , author Braren, R.F. , year 2020 . title Secure, privacy-preserving and federated machine learning in medical imaging . journal Nature Machine Intelligence volume 2 , pages 305--311

    2020

  19. [27]

    , author Blomberg, T

    author Kumarakulasinghe, N.B. , author Blomberg, T. , author Liu, J. , author Leao, A.S. , author Papapetrou, P. , year 2020 . title Evaluating local interpretable model-agnostic explanations on clinical machine learning classification models , in: booktitle 2020 IEEE 33rd Int...

    2020

  20. [28]

    , author Pritzel, A

    author Lakshminarayanan, B. , author Pritzel, A. , author Blundell, C. , year 2017 . title Simple and scalable predictive uncertainty estimation using deep ensembles . journal Advances in neural information processing systems volume 30 , pages 6402--6413

    2017

  21. [29]

    , author Harvey, H

    author Larson, D.B. , author Harvey, H. , author Rubin, D.L. , author Irani, N. , author Justin, R.T. , author Langlotz, C.P. , year 2021 . title Regulatory frameworks for development and evaluation of artificial intelligence--based diagnostic imaging algorithms: summary and r...

    2021

  22. [30]

    , author Schalekamp, S

    author van Leeuwen, K.G. , author Schalekamp, S. , author Rutten, M.J. , author van Ginneken, B. , author de Rooij, M. , year 2021 . title Artificial intelligence in radiology: 100 commercially available products and their scientific evidence . journal European radiology volum...

    2021

  23. [31]

    , author Meer, A

    author Makower, J. , author Meer, A. , author Denend, L. , year 2010 . title Fda impact on us medical technology innovation: a survey of over 200 medical technology companies . journal Arlington (Virginia): National Venture Capital Association

    2010

  24. [32]

    , author Wells, W.M

    author Mehrtash, A. , author Wells, W.M. , author Tempany, C.M. , author Abolmaesumi, P. , author Kapur, T. , year 2020 . title Confidence calibration and predictive uncertainty estimation for deep medical image segmentation . journal IEEE transactions on medical imaging volum...

    2020

  25. [33]

    , author Gonz \'a lez, C

    author Mehrtens, H.A. , author Gonz \'a lez, C. , author Mukhopadhyay, A. , year 2022 . title Improving robustness and calibration in ensembles with diversity regularization , in: booktitle DAGM German Conference on Pattern Recognition , organization Springer . pp. pages 36--50

    2022

  26. [34]

    , author Chakraborty, J

    author Midya, A. , author Chakraborty, J. , author G \"o nen, M. , author Do, R.K. , author Simpson, A.L. , year 2018 . title Influence of ct acquisition and reconstruction parameters on radiomic feature reproducibility . journal Journal of Medical Imaging volume 5 , pages 011020

    2018

  27. [35]

    , author Banerjee, O

    author Moor, M. , author Banerjee, O. , author Abad, Z.S.H. , author Krumholz, H.M. , author Leskovec, J. , author Topol, E.J. , author Rajpurkar, P. , year 2023 . title Foundation models for generalist medical artificial intelligence . journal Nature volume 616 , pages 259--265

    2023

  28. [36]

    title Notified body capacity and reviews

    author Open Regulatory , year 2023 . title Notified body capacity and reviews . howpublished https://openregulatory.com/notified-bodies/ . note Accessed: 2023-01-25

    2023

  29. [37]

    org, E.S

    author of Radiology (ESR) communications@ myesr. org, E.S. , year 2018 . title Esr paper on structured reporting in radiology . journal Insights into imaging volume 9 , pages 1--7

    2018

  30. [38]

    , author Hofmanninger, J

    author Perkonigg, M. , author Hofmanninger, J. , author Herold, C.J. , author Brink, J.A. , author Pianykh, O. , author Prosch, H. , author Langs, G. , year 2021 . title Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging . journal Na...

    2021

  31. [39]

    , author Silberzweig, J.E

    author Powell, D.K. , author Silberzweig, J.E. , year 2015 . title State of structured reporting in radiology, a survey . journal Academic radiology volume 22 , pages 226--233

    2015

  32. [40]

    , author Stammer, W

    author Schramowski, P. , author Stammer, W. , author Teso, S. , author Brugger, A. , author Herbert, F. , author Shao, X. , author Luigs, H.G. , author Mahlein, A.K. , author Kersting, K. , year 2020 . title Making deep neural networks right for the right scientific reasons by...

    2020

  33. [41]

    , author Han, K

    author Shin, H.J. , author Han, K. , author Ryu, L. , author Kim, E.K. , year 2023 . title The impact of artificial intelligence on the reading times of radiologists for chest radiographs . journal NPJ Digital Medicine volume 6 , pages 82

    2023

  34. [42]

    , author van Workum, F

    author Sluijter, C.E. , author van Workum, F. , author Wiggers, T. , author van de Water, C. , author Visser, O. , author van Slooten, H.J. , author Overbeek, L.I. , author Nagtegaal, I.D. , year 2019 . title Improvement of care in patients with colorectal cancer: influence of...

    2019

  35. [43]

    , author Severn, M

    author Smith, A. , author Severn, M. , year 2022 . title An overview of continuous learning artificial intelligence-enabled medical devices . journal Canadian Journal of Health Technologies volume 2

    2022

  36. [44]

    , author Cester, D

    author Van Timmeren, J.E. , author Cester, D. , author Tanadini-Lang, S. , author Alkadhi, H. , author Baessler, B. , year 2020 . title Radiomics in medical imaging—“how-to” guide and critical reflection . journal Insights into imaging volume 11 , pages 1--16

    2020

  37. [45]

    , author Von dem Bussche, A

    author Voigt, P. , author Von dem Bussche, A. , year 2017 . title The eu general data protection regulation (gdpr) . journal A Practical Guide, 1st Ed., Cham: Springer International Publishing volume 10 , pages 10--5555

    2017

  38. [46]

    , author Feuerriegel, S

    author Vokinger, K.N. , author Feuerriegel, S. , author Kesselheim, A.S. , year 2021 . title Continual learning in medical devices: Fda's action plan and beyond . journal The Lancet Digital Health volume 3 , pages e337--e338

    2021

  39. [47]

    , author Gasser, U

    author Vokinger, K.N. , author Gasser, U. , year 2021 . title Regulating ai in medicine in the united states and europe . journal Nature Machine Intelligence volume 3 , pages 738--739

    2021

  40. [48]

    , author Wu, K

    author Wu, E. , author Wu, K. , author Daneshjou, R. , author Ouyang, D. , author Ho, D.E. , author Zou, J. , year 2021 . title How medical ai devices are evaluated: limitations and recommendations from an analysis of fda approvals . journal Nature Medicine volume 27 , pages 582--584

    2021

  41. [49]

    , author Wu, E

    author Wu, K. , author Wu, E. , author Rodolfa, K. , author Ho, D.E. , author Zou, J. , year 2024 . title Regulating ai adaptation: An analysis of ai medical device updates , in: booktitle Conference on Health, Inference, and Learning , organization PMLR . pp. pages 477--488

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.