Pith. sign in

REVIEW 3 major objections 4 minor 45 references

Uncovering Neuroimaging Biomarkers of Brain Tumor Surgery with AI-Driven Methods

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A global explanation optimizer on paired pre/post-surgery MRI of 49 glioma patients identifies survival-relevant hubs in sensory and frontal cortex and outperforms six baseline XAI methods on faithfulness.

desk verdict Rare paired pre/post-surgery MRI dataset and a plausible pipeline, but the survival classifier is near chance, so the biomarker claims do not hold. read the letter →

arxiv 2507.04881 v2 pith:XEFTFC4I submitted 2025-07-07 eess.IV cs.CV

Carmen Jimenez-Mesa , Yizhou Wan , Guilio Sansone , Francisco J. Martinez-Murcia , Javier Ramirez , Pietro Lio , Juan M. Gorriz , Stephen J. Price
show 2 more authors
John Suckling Michail Mamalakis
This is my paper
classification eess.IVcs.CV
keywords braintumorsurgeryexplainableAIglobalexplanationoptimizersurvivalpredictionstructuralMRIPCAfeatureengineeringgliomaneuroimagingbiomarkers
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 tries to establish that paired pre- and post-surgery structural MRI from 49 glioma patients contains a survival signal that can be surfaced by combining PCA-based feature engineering with explainable deep learning. The authors claim that a global explanation optimizer they introduce produces cohort-level maps that are more faithful and less noisy than six state-of-the-art attribution methods. Clinically, they argue that surgery-driven changes in sensory and frontal hubs, especially early auditory cortex, insular and frontal opercular areas, and orbital and polar frontal cortex, distinguish shorter-term from longer-term survivors, pointing to regions worth preserving during resection. If true, the framework would turn scarce longitudinal imaging into actionable surgical guidance rather than just a diagnostic tool.

What carries the argument

The load-bearing mechanism is the global explanation optimizer: an encoder-decoder that takes three attribution maps, plus their weighted average, as inputs and is trained with a multi-objective loss combining Gini-index sparseness, faithfulness measured as the correlation between attribution sums and output changes under perturbation, and SSIM similarity between the optimized map and the first PCA component of the structural MRI. The PCA component serves as a structural reference, so the optimizer is explicitly steered toward explanations that match dominant anatomical variation. This design is what the paper credits for both the fidelity and the comprehensibility gains, and it is also what lets the pipeline output a single cohort-level map instead of method-dependent attributions.

What would settle it

Take the same 49 paired scans and retrain the survival classifier with the survival labels randomly permuted across patients; if the global explanation optimizer still produces high-faithfulness maps (near 0.91) and the same atlas regions, such as early auditory, insular/frontal opercular, and orbital/polar frontal cortex, pass the thresholding criteria, those biomarkers are not tied to true survival duration.

Watch

Extended reading notes

Core claim

On the paper's own terms, the core discovery is a two-phase pipeline: latent-space PCA maps of pre-to-post-surgical structural change, followed by a survival classifier whose explanations are refined by a global optimizer. The optimizer blends sparseness, faithfulness, and structural similarity to the first PCA component of the imaging data, and the paper reports that it achieves a faithfulness score of 0.913 with lower RMSE, MAE, and MSM than Gradient SHAP, Guided Backpropagation, Guided Grad-CAM, Input×Gradient, Integrated Gradients, and Kernel SHAP. The resulting global maps, thresholded against a standard cortical atlas, implicate early auditory, insular/frontal opercular, and orbital/polar frontal regions as hubs of reorganization, with longer-term survivors showing more frontal and cingulate engagement and shorter-term survivors showing diffuse posterior sensory and visual involvement.

Load-bearing premise

The explanation maps are only survival biomarkers if the survival classifier that produced them carries genuine signal, yet its 5-fold accuracy (0.67) and F1 (0.52) barely exceed the 0.65 accuracy of always predicting the majority class in the 49 patients.

Editorial extensions

If this is right

  • Surgical planning could prioritize sparing sensory and frontal hubs, such as early auditory and orbital/polar frontal cortex, if they reproducibly mark longer survival.
  • Cohort-level XAI maps from the optimizer can serve as candidate biomarkers for risk stratification in glioma patients.
  • Unsupervised pretraining on a larger public MRI cohort reduces the variance of downstream survival classification on the small paired dataset.
  • The framework can be extended to other outcome variables and imaging modalities, such as diffusion or functional MRI, using the same optimizer.
  • Contradictory attributions across XAI methods may be reconcilable by optimization rather than by arbitrary averaging.

Reading between the lines

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

  • Because the survival classifier's mean accuracy (0.67) and F1 (0.52) barely clear a majority-class baseline (0.65) on 49 patients, the biological reading of the explanation maps should be re-tested with label permutation or external validation before being applied in surgery planning.
  • The optimizer's similarity term ties explanations to the first PCA component, which captures dominant variance rather than survival-specific variance; a variant that aligns explanations to survival gradients instead would show whether the identified hubs are genuinely outcome-related.
  • A concrete next test is to measure regional volumes of the implicated hubs in an independent longitudinal cohort and fit a Cox proportional-hazards model, checking whether those volumes predict survival beyond age and tumor grade.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a two-phase framework for identifying survival-related neuroimaging biomarkers from paired pre- and post-surgery structural MRI of 49 glioma patients. Phase I uses PCA-based latent-space feature engineering to map structural variability between shorter- and longer-term survivors. Phase II trains a binary survival classifier (Swin-Unet or autoencoder encoder with MLP or attention head, frozen or fine-tuned), applies six XAI attribution methods, and introduces a "global explanation optimizer" that combines sparseness, faithfulness, and SSIM-to-first-PCA into a single loss. The authors report that the optimizer outperforms existing XAI methods on faithfulness and sparseness, and they interpret the resulting maps as identifying regions such as early auditory cortex, insular/frontal opercular cortex, and orbital/polar frontal cortex as survival-related. The central claims are that the framework identifies reliable survival biomarkers and that the optimizer advances XAI fidelity and comprehensibility.

Significance. If the claims were supported, the work would be valuable: a rare paired pre/post-surgery dataset, a pipeline from imaging to interpretable maps, and an explicit attempt to reduce inter-method XAI variability in a clinical setting. The dataset itself and the open-code commitment are strengths. However, the central biomarker claim rests on a survival classifier that is statistically indistinguishable from a majority-class baseline, and the optimizer evaluation is largely circular because the reported winning metrics are the same terms in its training loss. With the evidence as presented, the clinical conclusions about decision-making and emotional-regulation regions are not warranted, and the technical claim of advancing beyond state-of-the-art XAI is not established. The work is better framed as a hypothesis-generating methodology demonstration than as a validated biomarker discovery, and substantial revision would be required to support the current claims.

major comments (3)
  1. [Section 5.2, Fig. 4] The survival classifier that generates all explanations is not shown to be above chance. The best configuration (Swin-Unet with attention head, frozen encoder) has mean 5-fold CV accuracy 0.67, F1 0.52, precision 0.55, sensitivity 0.64 on 49 patients. A trivial majority-class predictor achieves 0.65 accuracy (32/49 longer-term). With n=49 and standard error of accuracy roughly 0.067, the 95% CI for the reported 0.67 includes 0.65 and 0.5. No permutation test, binomial test, or McNemar test against the majority baseline is reported. Since the subsequent biomarker maps are attributions of this classifier, and a chance-level model has no survival signal, the explanation maps cannot be interpreted as survival biomarkers without first demonstrating statistically significant predictive performance. This undercuts the abstract's claim that 'survival after oncological surgery is influenced by alterations in regions related to cognitive and sensory functions' and the surgical-guidance conclusions in Section 5.3.2 and the Discussion.
  2. [Section 3.2.4, Eq. (4); Table 3] The evaluation of the global explanation optimizer is circular. The loss in Eq. (4) explicitly includes faithfulness (Mfaith), sparseness (Msparse), and SSIM to the first PCA component of sMRI (losssim). Table 3 then reports that the optimizer achieves the highest faithfulness (0.913), favorable sparseness (0.537), and lowest RMSE/MAE/MSM relative to the same PCA target, and concludes it 'outperforms' other XAI methods. Because these metrics are part of the training objective, the comparison largely measures agreement with the objective on the training set, not generalizable explanation quality or an independent advantage over the baselines. To support the claim of advancing beyond state-of-the-art XAI, the optimizer should be evaluated on held-out explanations or on metrics that were not included in its loss, and comparisons should account for the fact that the baselines were not optimized for those criteria.
  3. [Section 5.1, Table 1; Section 5.3.2] The significance of the reported brain regions is determined by ad-hoc percentile thresholds (e.g., at least one voxel above the 95th percentile and 50% of voxels above the 80th percentile) without any multiple-comparison correction or null-model control. Given the voxel-wise nature of the analyses over high-dimensional images, such thresholds can easily yield spurious regions under the null. The paper should provide a permutation-based null distribution (e.g., shuffling survival labels or group assignments) to establish that the identified regions exceed what would be expected by chance, or explicitly label the findings as descriptive rather than inferential.
minor comments (4)
  1. [Section 3.2.3] The description of how the 'first PCA from local explanations' is computed is unclear: PCA is applied to the local explanation maps, but the text does not specify which data matrix (e.g., all voxels across subjects, per-condition) is used, nor how the first component is converted into the spatial maps shown in Fig. 5 and Table 1.
  2. [Section 4.1] The survival groups are defined by a fixed 10-month cutoff after the postoperative scan, but no sensitivity analysis is provided for this cutoff. The choice of 10 months is not motivated, and the robustness of the findings to different cutoffs should be discussed, especially given the small sample.
  3. [Table 3] The standard deviations in Table 3 are described as assessed across four global explanations (pre/post × shorter/longer-term), which is an extremely small sample for a standard deviation. This should be stated explicitly in the caption or text, and the interpretability of these error bars should be discussed.
  4. [Various] There are typographical and formatting issues: 'demostrate' in Fig. 4b caption, 'robustness' misspelled in Fig. 1, and the repeated 'diffusion MRI' in the Discussion. The paper would also benefit from a more careful definition of 'global explanation' versus 'cohort-level explanation' in the Methods.

Circularity Check

2 steps flagged · score 6.0 of 10

The optimizer's reported superiority is partly a restatement of its own loss function (Eq. 4 vs Table 3), and the optimizer design is imported from a same-author citation [6]; the survival-biomarker claim additionally rests on a classifier barely above majority chance.

  1. self definitional [Section 3.2.4 (Eq. 4) / Section 5.3.1 (Table 3)]
    "The cost function guiding the optimization integrates three key components: sparseness, as defined in [27]; faithfulness [28], to ensure consistency with model predictions; and similarity, to align the optimized explanation with a structural representation. ... losstotal(x,y)=l1 1/Mfaith(f,g;x)+l2 Msparse(f,g;x)+l3 losssim(x,y) (4) ... The proposed global explanation optimizer outperformed both the baseline explanation methods used during its training and testing ... In terms of faithfulness, the optimizer achieved a score of 0.913 ..."

    Equation (4) is the loss minimized to produce the optimized explanation; it contains exactly Mfaith, Msparse, and losssim, where losssim is SSIM against the first PCA component of the sMRI (Eq. 3). Table 3 and Section 5.3.1 then report that the optimizer 'wins' on faithfulness, sparseness, and closeness to that same PCA-derived target (RMSE/MAE/MSM). The evaluation is therefore a direct read-out of the training objective rather than an independent test. The abstract's technical claim that the optimizer 'advances beyond state-of-the-art XAI methods by enhancing both the fidelity and comprehensibility of model explanations' is consequently true by construction, since fidelity, sparseness, and PCA alignment are the very quantities the optimizer was fitted to maximize.

  2. self citation load bearing [Section 3.2.4; reference [6]]
    "we proposed a global explanation optimizer, building on the methodology introduced by [6] for optimizing explanation representations. Our framework follows the foundational design of the original approach, including a non-linear encoder-decoder architecture (Swin-Unet) and a multi-objective cost function."

    Reference [6] is Mamalakis et al., 'Solving the enigma: Enhancing faithfulness and comprehensibility in explanations of deep networks', 2025, a work by an overlapping author group (Michail Mamalakis is both a [6] author and a present-paper author). The paper's key technical contribution, the global explanation optimizer, is adopted from [6] rather than derived or independently verified here. The only explicitly claimed distinction is the evaluation strategy, and that evaluation strategy is itself the circular objective-based comparison described above. The central methodological novelty thus reduces to an unverified self-citation plus a metric that the loss function already optimizes.

full rationale

Two concrete circular steps were found. First, Eq. (4) defines the optimizer's training loss as a weighted combination of inverse faithfulness, sparseness, and SSIM against the first PCA component of the sMRI; Table 3 and Section 5.3.1 then report that the optimizer achieves the best faithfulness, sparseness, and PCA-alignment values. Because the evaluation metrics are the training objectives, the purported superiority of the global explanation optimizer is partly true by construction. Second, Section 3.2.4 states that the optimizer is built on the methodology of [6] and follows its foundational design; [6] is a 2025 same-author-group work, so the load-bearing methodological premise is a self-citation rather than an independent derivation in this manuscript. Separately, the biomarker conclusion is not circular but is statistically unsupported: Section 5.2 reports best 5-fold CV accuracy 0.67, F1 0.52, precision 0.55 on n=49, while a trivial majority-class predictor achieves 0.65 accuracy; no permutation, binomial, or McNemar test against that baseline is reported. The paper's stated limitation about sample size in Section 6 does not address this baseline comparison. Under the reviewing rules, this lack of demonstrated survival signal in the classifier belongs to correctness risk rather than circularity, but it materially weakens the abstract's clinical claim. The overall score is 6 because one central 'prediction' (optimizer superiority) reduces by construction and the optimizer design is self-citation-borne, while the biomarker claim itself is not equationally forced.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claims rest on several hand-set parameters (PCA component counts, survival cutoff, region thresholds, loss weights) and on domain assumptions about a weak classifier's informativeness. No new physical entities are introduced.

free parameters (5)
  • Number of PCA components = 8 (brain), 12 (tumor)
    Chosen to explain >80% variance, which is data-dependent and specific to this dataset.
  • Survival cutoff = 10 months
    Patients dying within 10 months of post-op scan labeled shorter-term; this threshold is arbitrary and affects class balance (17 vs 32).
  • Region significance thresholds = >=1 voxel above 95th percentile and >=50% voxels above 80th percentile; cosine: <5th percentile and volume 20th…
    Hand-set criteria used to convert continuous maps into region lists; no multiple comparison correction.
  • Loss weights l1, l2, l3 = 0.4, 0.3, 0.3
    Selected via hyperparameter tuning on the same dataset, so results are partly fit to the data.
  • Number of random perturbations for faithfulness = 20
    Fixed for faithfulness computation; the choice may affect the score.
assumptions (5)
  • domain assumption PCA of sMRI captures meaningful structural variation related to survival.
    Central to Phase I; no independent evidence that survival-relevant variance is in the first 8/12 components.
  • domain assumption Attribution maps from a near-chance classifier reflect biological mechanisms.
    Explanations are interpreted as biomarkers, but classifier accuracy is near the majority baseline.
  • domain assumption The 10-month dichotomization ignores censoring and is a valid survival outcome.
    Longer-term group includes patients alive at study end; no survival analysis is performed.
  • domain assumption Aggregating local explanations via PCA yields meaningful global explanations.
    No theoretical or empirical justification is provided for this aggregation strategy.
  • ad hoc to paper Percentile thresholds on explanation maps identify true brain regions without multiple-comparison correction.
    Thresholds are chosen post hoc and are not derived from any statistical model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncovering Neuroimaging Biomarkers of Brain Tumor Surgery with AI-Driven Methods." pith.science (2026). https://pith.science/paper/XEFTFC4I

@misc{pith2026250704881,
  author       = {Pith},
  title        = {Pith review of: Uncovering Neuroimaging Biomarkers of Brain Tumor Surgery with AI-Driven Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XEFTFC4I}},
  note         = {Machine review of arXiv:2507.04881}
}
read the original abstract

Brain tumor resection is a highly complex procedure with profound implications for survival and quality of life. Predicting patient outcomes is crucial to guide clinicians in balancing oncological control with preservation of neurological function. However, building reliable prediction models is severely limited by the rarity of curated datasets that include both pre- and post-surgery imaging, given the clinical, logistical and ethical challenges of collecting such data. In this study, we develop a novel framework that integrates explainable artificial intelligence (XAI) with neuroimaging-based feature engineering for survival assessment in brain tumor patients. We curated structural MRI data from 49 patients scanned pre- and post-surgery, providing a rare resource for identifying survival-related biomarkers. A key methodological contribution is the development of a global explanation optimizer, which refines survival-related feature attribution in deep learning models, thereby improving both the interpretability and reliability of predictions. From a clinical perspective, our findings provide important evidence that survival after oncological surgery is influenced by alterations in regions related to cognitive and sensory functions. These results highlight the importance of preserving areas involved in decision-making and emotional regulation to improve long-term outcomes. From a technical perspective, the proposed optimizer advances beyond state-of-the-art XAI methods by enhancing both the fidelity and comprehensibility of model explanations, thus reinforcing trust in the recognition patterns driving survival prediction. This work demonstrates the utility of XAI-driven neuroimaging analysis in identifying survival-related variability and underscores its potential to inform precision medicine strategies in brain tumor treatment.

Figures

Figures reproduced from arXiv: 2507.04881 by the authors.

Figure 1
Figure 1. Framework proposed. Imaging processing is followed by two different analysis. Phase I: in the feature [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Examples of hyperparameter tuning results for (a) unsupervised learning, (b) global explanation models [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Variability across severity conditions. Top: First PCA components analysis showing atlas-based Euclidean [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Examples of training and validation results in the ablation study of Swin-Unet and AutoEncoder variants (a) [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Explainability analysis results. For each severity condition (shorter-term and longer-term survivals), multi [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Surgical and brain regions with the most significant changes according to the combined framework for [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 38 canonical work pages

  1. [1]

    Dadario, Bledi Brahimaj, Jacky Yeung, and Michael E

    Nicholas B. Dadario, Bledi Brahimaj, Jacky Yeung, and Michael E. Sughrue. Reducing the cognitive footprint of brain tumor surgery.Frontiers in Neurology, 12, August 2021

    work page 2021

  2. [2]

    Perioperative and postoperative quality of life in patients with glioma–a longitudinal cohort study.World neurosurgery, 117:e465–e474, 2018

    Christina Drewes, Lisa Millgård Sagberg, Asgeir Store Jakola, and Ole Solheim. Perioperative and postoperative quality of life in patients with glioma–a longitudinal cohort study.World neurosurgery, 117:e465–e474, 2018

    work page 2018

  3. [3]

    Principal component analysis.Nature Reviews Methods Primers, 2(1):100, 2022

    Michael Greenacre, Patrick JF Groenen, Trevor Hastie, Alfonso Iodice d’Enza, Angelos Markos, and Elena Tuzhilina. Principal component analysis.Nature Reviews Methods Primers, 2(1):100, 2022

    2022

  4. [4]

    Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018

    arXiv 2018

  5. [5]

    Computational approaches to explainable artificial intelligence: advances in theory, applica- tions and trends.Information Fusion, 100:101945, 2023

    Juan M Górriz et al. Computational approaches to explainable artificial intelligence: advances in theory, applica- tions and trends.Information Fusion, 100:101945, 2023

    work page 2023

  6. [6]

    Solving the enigma: Enhancing faithfulness and comprehensibility in explanations of deep networks, 2025

    Michail Mamalakis et al. Solving the enigma: Enhancing faithfulness and comprehensibility in explanations of deep networks, 2025

    work page 2025

  7. [7]

    Christiaan HB Van Niftrik et al. Machine learning algorithm identifies patients at high risk for early complications after intracranial tumor surgery: registry-based cohort study.Neurosurgery, 85(4):E756–E764, 2019

    work page 2019

  8. [8]

    Routine ICU surveillance after brain tumor surgery: Patient selection using machine learning.Journal of Clinical Medicine, 13(19):5747, 2024

    Jan-Oliver Neumann, Stephanie Schmidt, Amin Nohman, Paul Naser, Martin Jakobs, and Andreas Unterberg. Routine ICU surveillance after brain tumor surgery: Patient selection using machine learning.Journal of Clinical Medicine, 13(19):5747, 2024

    work page 2024

Show all 45 references
  1. [9]

    Predicting inpatient length of stay after brain tumor surgery: developing machine learning ensembles to improve predictive performance

    Whitney E Muhlestein, Dallin S Akagi, Jason M Davies, and Lola B Chambless. Predicting inpatient length of stay after brain tumor surgery: developing machine learning ensembles to improve predictive performance. Neurosurgery, 85(3):384–393, 2019

    2019

  2. [10]

    Evaluation and classification of the brain tumor mri using machine learning technique.Journal of Control Engineering and Applied Informatics, 21(4):12–21, 2019

    R Pugalenthi, MP Rajakumar, J Ramya, and V Rajinikanth. Evaluation and classification of the brain tumor mri using machine learning technique.Journal of Control Engineering and Applied Informatics, 21(4):12–21, 2019

    2019

  3. [11]

    A deep learning approach for brain tumor classification and segmentation using a multiscale convolutional neural network.Healthcare, 9(2):153, February 2021

    Francisco Javier Díaz-Pernas, Mario Martínez-Zarzuela, Míriam Antón-Rodríguez, and David González-Ortega. A deep learning approach for brain tumor classification and segmentation using a multiscale convolutional neural network.Healthcare, 9(2):153, February 2021

    2021

  4. [12]

    Pre and post operative brain tumor segmentation and classification for prolonged survival

    P Sobha Xavier, G Raju, and SU Asawthy. Pre and post operative brain tumor segmentation and classification for prolonged survival. InInternational Conference on Soft Computing and Pattern Recognition, pages 608–616. Springer, 2021

    2021

  5. [13]

    Jakub Nalepa et al. Deep learning automates bidimensional and volumetric tumor burden measurement from mri in pre-and post-operative glioblastoma patients.Computers in biology and medicine, 154:106603, 2023

    2023

  6. [14]

    Machine learning for grading prediction and survival analysis in high grade glioma

    Xiangzhi Li, Xueqi Huang, Yi Shen, Sihui Yu, Lin Zheng, Yunxiang Cai, Yang Yang, Renyuan Zhang, Lingying Zhu, and Enyu Wang. Machine learning for grading prediction and survival analysis in high grade glioma. Scientific Reports, 15(1):16955, 2025

    2025

  7. [15]

    Machine learning and neurosurgical outcome prediction: a systematic review.World neurosurgery, 109:476–486, 2018

    Joeky T Senders et al. Machine learning and neurosurgical outcome prediction: a systematic review.World neurosurgery, 109:476–486, 2018

    2018

  8. [16]

    Machine learning for predicting post-operative outcomes in meningiomas: a systematic review and meta-analysis.Acta Neurochirurgica, 166(1):1–14, 2024

    Siraj Y Abualnaja, James S Morris, Hamza Rashid, William H Cook, and Adel E Helmy. Machine learning for predicting post-operative outcomes in meningiomas: a systematic review and meta-analysis.Acta Neurochirurgica, 166(1):1–14, 2024. 16 APREPRINT- SEPTEMBER15, 2025

    2024

  9. [17]

    Imaging surrogates of infiltration obtained via multiparametric imaging pattern analysis predict subsequent location of recurrence of glioblastoma

    Hamed Akbari, Luke Macyszyn, Xiao Da, Michel Bilello, Ronald L Wolf, Maria Martinez-Lage, George Biros, Michelle Alonso-Basanta, Donald M O’Rourke, and Christos Davatzikos. Imaging surrogates of infiltration obtained via multiparametric imaging pattern analysis predict subsequ...

    2016

  10. [18]

    Latent correlation representation learning for brain tumor segmentation with missing mri modalities.IEEE Transactions on Image Processing, 30:4263–4274, 2021

    Tongxue Zhou, Stéphane Canu, Pierre Vera, and Su Ruan. Latent correlation representation learning for brain tumor segmentation with missing mri modalities.IEEE Transactions on Image Processing, 30:4263–4274, 2021

    2021

  11. [19]

    The explanation necessity for healthcare ai, 2024

    Michail Mamalakis, Héloïse de Vareilles, Graham Murray, Pietro Lio, and John Suckling. The explanation necessity for healthcare ai, 2024

    2024

  12. [20]

    Explainable artificial intelligence (xai) 2.0: A manifesto of open challenges and interdisciplinary research directions.Information Fusion, 106:102301, 2024

    Luca Longo et al. Explainable artificial intelligence (xai) 2.0: A manifesto of open challenges and interdisciplinary research directions.Information Fusion, 106:102301, 2024

    2024

  13. [21]

    Explainable deep learning framework for brain tumor detection: Integrating lime, grad-cam, and shap for enhanced accuracy.Medical Engineering & Physics, page 104405, 2025

    Abdurrahim Akgündo˘gdu and ¸ Serife Çelikba¸ s. Explainable deep learning framework for brain tumor detection: Integrating lime, grad-cam, and shap for enhanced accuracy.Medical Engineering & Physics, page 104405, 2025

    2025

  14. [22]

    Roth, and Daguang Xu

    Ali Hatamizadeh, Vishwesh Nath, Yucheng Tang, Dong Yang, Holger R. Roth, and Daguang Xu. Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images. In Alessandro Crimi and Spyridon Bakas, editors,Brainlesion: Glioma, Multiple Sclerosis, Stroke and T...

    2022

  15. [23]

    Striving for simplicity: The all convolutional net, 2015

    Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin Riedmiller. Striving for simplicity: The all convolutional net, 2015

    2015

  16. [24]

    Grad-cam: Visual explanations from deep networks via gradient-based localization.International Journal of Computer Vision, 128(2):336–359, October 2019

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization.International Journal of Computer Vision, 128(2):336–359, October 2019

    2019

  17. [25]

    A unified approach to interpreting model predictions

    Scott M Lundberg and Su-In Lee. A unified approach to interpreting model predictions. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017

    2017

  18. [26]

    Axiomatic attribution for deep networks

    Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic attribution for deep networks. InProceedings of the 34th International Conference on Machine Learning (ICML), pages 3319–3328. PMLR, 2017

    2017

  19. [27]

    Concise explanations of neural networks using adversarial training

    Prasad Chalasani, Jiefeng Chen, Amrita Roy Chowdhury, Xi Wu, and Somesh Jha. Concise explanations of neural networks using adversarial training. InInternational conference on machine learning, pages 1383–1391. PMLR, 2020

    2020

  20. [28]

    On the (in) fidelity and sensitivity of explanations.Advances in neural information processing systems, 32, 2019

    Chih-Kuan Yeh, Cheng-Yu Hsieh, Arun Suggala, David I Inouye, and Pradeep K Ravikumar. On the (in) fidelity and sensitivity of explanations.Advances in neural information processing systems, 32, 2019

    2019

  21. [29]

    Quantus: An explainable ai toolkit for responsible evaluation of neural network explanations and beyond.Journal of Machine Learning Research, 24(34):1–11, 2023

    Anna Hedström et al. Quantus: An explainable ai toolkit for responsible evaluation of neural network explanations and beyond.Journal of Machine Learning Research, 24(34):1–11, 2023

    2023

  22. [30]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 13(4):600–612, 2004

    2004

  23. [31]

    Automated brain extraction of multisequence mri using artificial neural networks.Human brain mapping, 40(17):4952–4964, 2019

    Fabian Isensee et al. Automated brain extraction of multisequence mri using artificial neural networks.Human brain mapping, 40(17):4952–4964, 2019

    2019

  24. [32]

    The 2024 brain tumor segmentation (brats) challenge: glioma segmentation on post-treatment mri.arXiv preprint arXiv:2405.18368, 2024

    Maria Correia de Verdier, Rachit Saluja, Louis Gagnon, Dominic LaBella, Ujjwall Baid, Nourel Hoda Tahon, Martha Foltyn-Dumitru, Jikai Zhang, Maram Alafif, Saif Baig, et al. The 2024 brain tumor segmentation (brats) challenge: glioma segmentation on post-treatment mri.arXiv pre...

    2024 arXiv

  25. [33]

    A multi-modal parcellation of human cerebral cortex.Nature, 536(7615):171–178, 2016

    Matthew F Glasser et al. A multi-modal parcellation of human cerebral cortex.Nature, 536(7615):171–178, 2016

    2016

  26. [34]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2014

    2014

  27. [35]

    Umang Bhatt, Adrian Weller, and José M. F. Moura. Evaluating and aggregating feature-based model explanations. 2020

    2020

  28. [36]

    Concise explanations of neural networks using adversarial training

    Prasad Chalasani, Jiefeng Chen, Amrita Roy Chowdhury, Xi Wu, and Somesh Jha. Concise explanations of neural networks using adversarial training. In Hal Daumé III and Aarti Singh, editors,Proceedings of the 37th International Conference on Machine Learning (ICML), volume 119 of...

    2018 arXiv

  29. [37]

    Effect of brain surgery on auditory and motor cortex activation: a preliminary functional magnetic resonance imaging study.Neurosurgery, 57(2):249–256, 2005

    Salla-Maarit Kokkonen, Vesa Kiviniemi, Minna Mäkiranta, Sanna Yrjänä, John Koivukangas, and Osmo Tervonen. Effect of brain surgery on auditory and motor cortex activation: a preliminary functional magnetic resonance imaging study.Neurosurgery, 57(2):249–256, 2005. 17 APREPRINT...

    2005

  30. [38]

    The influence of frontal lobe tumors and surgical treatment on advanced cognitive functions.World Neurosurgery, 91:340–346, 2016

    Shengyu Fang, Yinyan Wang, and Tao Jiang. The influence of frontal lobe tumors and surgical treatment on advanced cognitive functions.World Neurosurgery, 91:340–346, 2016

    2016

  31. [39]

    Primary cognitive factors impaired after glioma surgery and associated brain regions.Behavioural neurology, 2020(1):7941689, 2020

    Chiharu Niki, Takatsune Kumada, Takashi Maruyama, Manabu Tamura, Takakazu Kawamata, and Yoshihiro Muragaki. Primary cognitive factors impaired after glioma surgery and associated brain regions.Behavioural neurology, 2020(1):7941689, 2020

    2020

  32. [40]

    Reward- related reversal learning after surgical excisions in orbito-frontal or dorsolateral prefrontal cortex in humans

    J Hornak, J O’doherty, Jessica Bramham, Edmund T Rolls, Robin G Morris, PR Bullock, and CE Polkey. Reward- related reversal learning after surgical excisions in orbito-frontal or dorsolateral prefrontal cortex in humans. Journal of cognitive neuroscience, 16(3):463–478, 2004

    2004

  33. [41]

    Riho Nakajima, Masashi Kinoshita, Hirokazu Okita, Tetsutaro Yahata, and Mitsutoshi Nakada. Glioma surgery under awake condition can lead to good independence and functional outcome excluding deep sensation and visuospatial cognition.Neuro-Oncology Practice, 6(5):354–363, 2019

    2019

  34. [42]

    Auditory cortical plasticity: a comparison with other sensory systems.Trends in neuro- sciences, 22(2):74–80, 1999

    Josef P Rauschecker. Auditory cortical plasticity: a comparison with other sensory systems.Trends in neuro- sciences, 22(2):74–80, 1999

    1999

  35. [43]

    The insular lobe and brain plasticity: lessons from tumor surgery.Clinical neurology and neurosurgery, 108(6):543–548, 2006

    Hugues Duffau, Luc Taillandier, Peggy Gatignol, and Laurent Capelle. The insular lobe and brain plasticity: lessons from tumor surgery.Clinical neurology and neurosurgery, 108(6):543–548, 2006

    2006

  36. [44]

    Neuroplasticity mechanisms in frontal brain gliomas: a preliminary study.Frontiers in Neurology, 13:867048, 2022

    Micaela Mitolo, Matteo Zoli, Claudia Testa, Luca Morandi, Magali Jane Rochat, Fulvio Zaccagna, Matteo Martinoni, Francesca Santoro, Sofia Asioli, Filippo Badaloni, et al. Neuroplasticity mechanisms in frontal brain gliomas: a preliminary study.Frontiers in Neurology, 13:867048, 2022

    2022

  37. [45]

    What do we know about pre-and postoperative plasticity in patients with glioma? a review of neuroimaging and intraoperative mapping studies.NeuroImage: Clinical, 28:102435, 2020

    Elisa Cargnelutti, Tamara Ius, Miran Skrap, and Barbara Tomasino. What do we know about pre-and postoperative plasticity in patients with glioma? a review of neuroimaging and intraoperative mapping studies.NeuroImage: Clinical, 28:102435, 2020. 18

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.