HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification

Marwan Ait Haddou; Mohamed Bennai

REVIEW 5 major objections 6 minor 40 references

HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification

T0 review · 5 major / 6 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 The paper claims a hybrid quantum-classical model with a 5-qubit trainable layer achieves 96.48% accuracy on four-class brain-tumor MRI and localizes tumors better than its classical counterpart.

desk verdict Central comparison is uninterpretable: classical baseline undefined, attention evaluation circular, and confusion matrices don't match reported accuracy. read the letter →

arxiv 2506.21937 v1 pith:CBWJWWHK submitted 2025-06-27 cs.LG

Marwan Ait Haddou , Mohamed Bennai This is my paper

classification cs.LG MSC 68T0768Q1281P68

keywords hybridquantum-classicalmodelbraintumorclassificationquantummachinelearningattentionmechanismMRIconsistencylossinterpretabilityamplitudeencoding

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

The paper claims that adding a small trainable quantum layer to a convolutional brain-tumor classifier materially improves both accuracy and interpretability. On 7,576 MRI scans spanning normal, meningioma, glioma, and pituitary classes, the hybrid model reports 96.48% test accuracy versus 86.72% for its classical counterpart, with the largest gain in glioma recall (0.98 vs. 0.69). It further reports that the quantum model's attention maps align better with ground-truth tumor masks at high-confidence thresholds, measured by a Jaccard index at $\tau=0.99$ (0.43220 vs. 0.40219, $p=0.0017$). The authors attribute the gain to quantum feature separability visible in t-SNE projections and to attention-consistency training. A sympathetic reading is that quantum processing plus attention supervision is a viable path for clinically meaningful medical-image classification.

What carries the argument

The load-bearing component is the parameterized quantum circuit layer: attention-refined feature maps are flattened and projected to a vector $z \in \mathbb{R}^{2q \cdot c}$, amplitude-embedded into $c=5$ parallel circuits with $q=5$ qubits each, evolved through $d=2$ layers of $R_z$-$R_y$-$R_z$ rotations interlaced with CNOT gates, and read out via Pauli-$Y$ expectation values; the concatenated outputs feed a fully connected classifier. This provides a compact, trainable, nonlinear feature transformation in Hilbert space. A second mechanism is the dual attention module adapted from the convolutional block attention module, with channel attention followed by spatial attention and a residual spatial gate $F_{\mathrm{final}} = F \odot (1 + T_s)$, together with an attention-consistency loss combining BCE and Dice terms that steers the model's focus toward tumor masks.

What would settle it

Run the identical pipeline with the quantum layer replaced by a classical layer of comparable parameter count and nonlinearity, keeping the attention module and attention-consistency loss fixed; if the accuracy gap and Jaccard-index gap disappear or reverse, the quantum layer is not the cause. A simpler check is to report the architecture and training budget of the Classical Counterpart: if it omits attention or the attention-consistency loss, the paper's central attribution fails.

Watch

Extended reading notes

Core claim

The paper's central discovery claim is that replacing part of the classical processing head with a trainable quantum layer, while supervising attention maps against tumor masks, yields a brain-tumor classifier that is both more accurate and more localizable than the classical baseline. Test-set results show 96.48% overall accuracy and macro F1 0.96 for HQCM-EBTC versus 86.72% accuracy and macro F1 0.88 for the Classical Counterpart. The largest class-level improvement is in glioma recall, which rises from 0.69 to 0.98. The paper further claims that quantum-space features, visualized with t-SNE, form more compact and separated clusters, and that attention maps achieve higher Jaccard indices against ground-truth tumor masks at $\tau=0.99$ and $\tau=0.90$, with $p$-values 0.0017 and 0.0350.

Load-bearing premise

The claim rests on the classical counterpart being a fair control: if that baseline lacks the attention module or the attention-consistency loss, then the reported accuracy and localization gaps could come from those components rather than from quantum processing.

Editorial extensions

If this is right

A 5-qubit, depth-2 trainable quantum layer can lift in-distribution classification accuracy on four-class brain-tumor MRI by about 10 points over the stated classical baseline, with the largest gain in the hardest class, glioma.
Attention maps trained against tumor masks localize better at high confidence thresholds, making the model a candidate for weakly supervised tumor localization, not just classification.
The composite loss of classification error plus attention consistency gives a general recipe for steering hybrid models toward clinically relevant regions.
If confirmed, the result suggests that quantum feature spaces can provide separability that a classical fully connected layer of comparable size does not, motivating quantum layers in other high-dimensional medical imaging tasks.

Reading between the lines

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

Beyond the paper, the reported comparison cannot separate the quantum layer's contribution from the attention-consistency loss, so a reader should not infer that the 9.76-point accuracy gap is due to quantum processing alone until the classical control matches on both components.
Beyond the paper, because the circuit is simulated on a GPU and uses only 5 qubits at depth 2, the result carries no claim of quantum computational advantage; it is best read as evidence about a particular trainable feature map, and replacing the circuit with a fixed nonlinear random feature map of the same output dimension would test whether trainability or entanglement matters.
Beyond the paper, the attention-consistency loss is task-agnostic, so the same hybrid recipe could transfer to other lesion-localization problems, where the added value would be measurable as an increase in Jaccard index at high thresholds.

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.

Desk Editor's Note

read the letter

Dear Colleague,

The one thing to know: the paper's central claim—HQCM-EBTC beats its "Classical Counterpart" by 9.76 points—can't be evaluated, because the classical baseline is never defined. The text gives hyperparameters for the hybrid model only; there's no description of the control architecture, loss, augmentation, or training budget. That single omission makes the headline result uninterpretable.

What's actually new and worth crediting: the specific combination of CNN + multi-scale spatial/channel attention + five parallel 5-qubit amplitude-encoded circuits with an attention-consistency loss is not in the cited literature. The pipeline is described clearly, and the row and column totals in the confusion matrices are internally consistent, which suggests some care in reporting. The attention overlays in Figure 10 are a nice visual.

The soft spots, in rough order of severity. First, the Jaccard interpretability result is partially circular: Section 3.4 trains attention maps against ground-truth masks with BCE+Dice, and Section 4.6 scores them against those same masks. A high IoU at high thresholds mostly shows the auxiliary loss is working, not that the quantum layer helps localization. Second, the accuracy numbers don't line up with the confusion matrices: the classical matrix in Figure 9b sums to 996/1137 = 87.6% correct, not 86.72%; the hybrid matrix sums to 1098/1137 = 96.6%, not 96.48%. The gap is a few counts, but for a paper whose whole point is a 9.76-point gap, it matters. Third, Figure 7 labels the model "Brain-QANTIC," a name that never appears in the text. Fourth, the t-SNE comparison in Figure 8 compares an intermediate quantum feature space to the classical model's output layer, which isn't apples-to-apples. And no error bars, repeated runs, code, or data are provided.

The larger issue is that the accuracy gap could be entirely due to the attention module or the auxiliary loss, not the quantum layer. Prior cited work already exceeds 99% accuracy on similar tasks, so the 96.48% result isn't state of the art.

Who is this for? Someone wanting a concrete example of a hybrid QML pipeline might learn from the architecture description. But the empirical claims aren't reliable as written. I wouldn't send this to peer review in its current form; it needs a properly matched classical control, standard error reporting, an explanation of the "Brain-QANTIC" label, and code/data release. Without those, the quantum advantage is uninterpretable.

Best, [Your name]

Referee Report

5 major / 6 minor

Summary. The manuscript proposes HQCM-EBTC, a hybrid quantum-classical CNN that combines classical convolutional feature extraction, a CBAM-style dual-attention module, and a simulated quantum processing layer (5 qubits, depth 2, 5 parallel circuits) for four-class brain tumor classification on 7,576 MRI scans. The central empirical claims are (i) test accuracy of 96.48% versus 86.72% for an unspecified 'Classical Counterpart' (Table 1), (ii) improved per-class precision, recall, and F1 scores, (iii) better feature separability in t-SNE projections, and (iv) higher Jaccard alignment between attention maps and ground-truth tumor masks at high thresholds (Table 2). The paper concludes that quantum-enhanced processing improves both classification accuracy and interpretability.

Significance. The question addressed is timely: whether a small simulated quantum layer can add measurable value in medical image classification while preserving interpretability. The paper reports independent test-set metrics and provides confusion matrices, training curves, t-SNE visualizations, and a statistical test for the Jaccard comparison, which are appropriate empirical artifacts. If the 9.76-point accuracy gap were established against a well-matched classical baseline with repeated runs and a correct quantum circuit description, the result would be a useful contribution to quantum machine learning for medical imaging. As it stands, however, the missing baseline definition, an apparent dimensional inconsistency in the amplitude-embedding step, and internal inconsistencies between the reported accuracies and the confusion matrices prevent the current evidence from supporting the central claim. The work is potentially salvageable with a properly controlled ablation and a corrected manuscript.

major comments (5)

[§4.4, Table 1; §4.2, Table 3] The 'Classical Counterpart' used throughout the comparison is never specified. The manuscript does not state whether this baseline uses the same dual-attention module (§3.2), the same attention-consistency loss (§3.4), the same data augmentation, or the same training budget; Table 3 lists hyperparameters only for HQCM-EBTC. Because the attention module and the supervised attention loss can independently improve both accuracy and localization, the reported 9.76-point accuracy gap cannot be attributed to the quantum layer. In addition, all metrics come from a single train/validation/test split with no repeated runs or confidence intervals, so the stability of the 96.48% figure is unknown. A matched ablation with the quantum layer removed and multiple random seeds is required before the central claim can be evaluated.
[§3.3, amplitude-embedding description and Fig. 4] The quantum processing layer is dimensionally inconsistent as written. The flattened attention features are projected to a vector z ∈ R^{2q·c}, and with q=5 and c=5 this is a 50-dimensional vector, yet each of the c circuits is said to amplitude-encode a sub-vector into q=5 qubits, which requires 2^q = 32 amplitudes per circuit. The displayed state |ψ⟩ = Σ_{i=1}^{2q} z_i |i⟩ also uses the wrong upper limit unless 2q is a typo for 2^q. Either the projection dimension should be 2^q·c, or the encoding is not amplitude embedding; as written, the construction cannot be implemented. The authors should correct the dimension and provide the exact circuit decomposition or a code snippet.
[Table 1 vs. Fig. 9] The confusion matrices in Figure 9 are inconsistent with the overall accuracies in Table 1. Summing the test-set counts in Figure 9(a) gives a total of 1,137 samples and 1,098 correct predictions, i.e., 96.57% accuracy, not 96.48%; summing Figure 9(b) gives 996 correct predictions, i.e., 87.60%, not 86.72%. The discrepancy for the classical model corresponds to about ten samples and cannot be explained by rounding alone. The authors should reconcile the reported accuracy with the displayed confusion matrices or state explicitly if different test sets were used.
[§4.6, Table 2; §3.4, Eq. (8)–(9)] The Jaccard-index comparison is partially circular. The attention maps are trained against ground-truth tumor masks through the attention-consistency loss L_attn (BCE + Dice), and the same masks are then used to compute the Jaccard scores in Table 2. High Jaccard values at τ=0.99 therefore largely reflect that the auxiliary loss is doing its job, not that the quantum layer improves localization. The comparison would be informative only if the classical baseline also used identical attention-consistency supervision, which is not stated anywhere. The authors should either evaluate interpretability with an independent measure or explicitly restrict the claim to the effect of attention-consistency training.
[§4.3, Fig. 7; §4.4, Fig. 8] The t-SNE comparison in Figure 8 contrasts the quantum processing layer output with the fully connected layer output of the classical model, but since the classical model is not matched, the improved cluster separability may stem from the attention module, the attention-consistency loss, or the different layer being visualized. Moreover, Figure 7 labels the proposed model as 'Brain-QANTIC' while the text calls it HQCM-EBTC, making it difficult to verify which curves correspond to which model. The authors should provide a matched-feature comparison with the same layer depth and the same loss, and should unify the model name.

minor comments (6)

[§4.2 and Table 3] The text in Section 4.2 states a learning rate of 0.01, while Table 3 reports 0.011; please reconcile these values.
[§3.4, Eq. (8) and Table 3] Equation (8) defines L_attn = ζ L_BCE + γ L_Dice, but Table 3 labels ζ as 'Dice Loss Scaling' and γ as 'BCE Loss Scaling'; the labels and the text's description of ζ and γ should be made consistent.
[§4.1] Section 4.1 does not state which of the three data sources provide ground-truth tumor masks or how the masks were aligned to the resized 128×128 images; this information is necessary to reproduce the attention-consistency loss and the Jaccard evaluation.
[§4.6, Table 2 and Fig. 10] The per-image Jaccard values in Figure 10 (e.g., 0.3865, 0.2013, 0.1159) are much lower than the mean values in Table 2 at the same threshold; please clarify whether Table 2 reports test-set means and how the example rows in Figure 10 were selected.
[§2 and §5] Section 2 claims enhanced out-of-distribution generalizability, but no out-of-distribution experiment is reported in Section 4; either add such an evaluation or remove the claim. The limitations discussion in Section 5 acknowledges overfitting risk but does not address the missing matched baseline or the single-split evaluation.
[References] Reference [35] is dated 'Accessed: 2025-11-20', which is after the manuscript's arXiv submission date of 27 June 2025; please correct the access date.

Circularity Check

1 steps flagged · score 6.0 of 10

Localization 'prediction' is the attention-consistency training target relabeled: Jaccard scores reuse the ground-truth masks used in Lattn; the accuracy claim is independently test-set grounded, but the quantum-attribution and classical baseline are unspecified.

fitted input called prediction [Section 3.4 (Training and Loss Function) and Section 4.6 (Attention Map Analysis), Table 2 / Figure 10]
"In addition, to ensure that the network's attention maps corresponded closely with ground-truth tumor masks, we incorporated an attention consistency loss Lattn. This loss combined binary cross-entropy (BCE) and Dice losses [7]: LBCE = − 1/B Σ [Ti log Ai + (1 − Ti) log(1 − Ai)] ... LDice = 1 − 2 (Σ TiAi + ε)/(Σ Ti + Σ Ai + ε). To quantify the alignment, we compute the Jaccard Index (IoU) at three thresholds: τ = 0.99, τ = 0.90, and τ = 0.75, using the formula: J(Aτ , M) = |Aτ ∩ M| / |Aτ ∪ M|, where Aτ is the binarized attention map at threshold τ and M is the ground-truth mask."

Section 3.4 explicitly trains the predicted attention map A against the ground-truth tumor mask T through BCE and Dice terms inside Lattn. Section 4.6 then reports the 'alignment' of the attention map with the same ground-truth masks via the Jaccard index. The reported values (0.43220 vs 0.40219 at tau=0.99; p=0.0017) therefore largely measure how well the supervised auxiliary loss was optimized, not an independent prediction of tumor localization. The metric is the training target relabeled as a result, so the localization advantage attributed to the quantum layer is partially forced by construction. Because the 'Classical Counterpart' is never specified as sharing or omitting the dual-attention module and Lattn, this Jaccard gap also cannot be cleanly attributed to quantum processing.

full rationale

The headline accuracy comparison (96.48% vs. 86.72%, Table 1) is a held-out test-set measurement and is not circular: it does not reduce to the model definition or to a fitted parameter. The classification metrics, confusion matrices, and t-SNE plots are external benchmarks. The genuinely circular element is the interpretability/localization claim. Equations in Section 3.4 train the attention map A to match the ground-truth tumor mask T via BCE and Dice losses, and Section 4.6 evaluates 'alignment' by computing Jaccard between the binarized attention map and the same ground-truth mask M. This is a fit-quality check presented as a quantum-localization prediction, so the Jaccard advantage at tau=0.99 and tau=0.90 is partially forced by Lattn. The undefined 'Classical Counterpart' is a serious control/comparability weakness, but it is a correctness and attribution concern rather than a derivation-circularity concern, so it does not further raise the circularity score. The only self-citation, reference [19] (co-author Bennai), appears in the literature review as an example of prior QCNN accuracy and is not load-bearing for any derivation in this paper. Since one of the two central claims (explainable localization) reduces to the supervised training objective while the other (accuracy) is independently grounded, the appropriate score is 6: partial circularity, not total.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central accuracy result rests on hand-chosen hyperparameters, an unspecified classical control, and an assumed-clean merged dataset. The localization result rests on attention maps that the loss function is explicitly trained to match. No code or data accompany the paper, so the ledger is dominated by unverified assumptions rather than invented entities.

free parameters (4)

Composite loss weights (alpha, beta, zeta, gamma) = 1, 1, 0.3, 0.7
Chosen by hand; control the trade-off between classification and attention-consistency objectives. No sensitivity analysis is reported.
Quantum circuit configuration (qubits, depth, circuits) = 5 qubits, depth 2, 5 parallel circuits
Chosen without ablation or justification; central to the claim that the quantum layer helps.
Learning rate and weight decay = 0.01, 1e-4
Chosen by hand; no sensitivity analysis is reported.
Cross-entropy class weights w_i = not reported
Defined in L_class but values are never given, and they affect training balance and the reported metrics.

assumptions (4)

ad hoc to paper The Classical Counterpart is a matched control: same data, augmentation, loss, and training schedule, with only the quantum layer removed.
The superiority claim in Table 1 requires this, but the paper never specifies the control architecture.
domain assumption Merged dataset sources are compatible, and labels and masks are correct with no patient-level leakage.
Accuracy and Jaccard claims depend on clean labels and split independence; no patient IDs or preprocessing code are provided.
domain assumption The spatial attention map Ts is a valid explanation of model decisions.
The interpretability claim treats attention as explanation; it is not validated against gradient-based or occlusion methods.
standard math PennyLane's amplitude embedding and expectation measurements correctly realize the stated quantum operations.
Simulation fidelity is assumed; no verification against analytic results or hardware is reported.

how reviews work

0 comments

Cite this review

Pith. "Pith review of HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification." pith.science (2026). https://pith.science/paper/CBWJWWHK

@misc{pith2026250621937,
  author       = {Pith},
  title        = {Pith review of: HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CBWJWWHK}},
  note         = {Machine review of arXiv:2506.21937}
}

read the original abstract

We propose HQCM-EBTC, a hybrid quantum-classical model for automated brain tumor classification using MRI images. Trained on a dataset of 7,576 scans covering normal, meningioma, glioma, and pituitary classes, HQCM-EBTC integrates a 5-qubit, depth-2 quantum layer with 5 parallel circuits, optimized via AdamW and a composite loss blending cross-entropy and attention consistency. HQCM-EBTC achieves 96.48% accuracy, substantially outperforming the classical baseline (86.72%). It delivers higher precision and F1-scores, especially for glioma detection. t-SNE projections reveal enhanced feature separability in quantum space, and confusion matrices show lower misclassification. Attention map analysis (Jaccard Index) confirms more accurate and focused tumor localization at high-confidence thresholds. These results highlight the promise of quantum-enhanced models in medical imaging, advancing both diagnostic accuracy and interpretability for clinical brain tumor assessment.

Figures

Figures reproduced from arXiv: 2506.21937 by the authors.

**Figure 1.** Overall architecture of the proposed HQCM-EBTC model. Data preprocessing prepares input images x ∈ R 1×H×W . Classical convolutional blocks extract a feature tensor F ∈ R 32× H 8 × W 8 , which is refined by the attention mechanism to produce Ffinal. This refined tensor is flattened and projected into z ∈ R 2 q×c , then processed by the Quantum Processing Layer via amplitude embedding, entangling gates, and measureme… view at source ↗

**Figure 2.** Illustration of the classical feature extraction pipeline. The input image (left) was suc [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Visualization of the dual-attention module. The top branch (channel attention) aggre [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Schematic of the quantum processing layer. The classical feature vector was split and [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 6.** Figure 6: Sample images from the integrated dataset, illustrating normal brain structures and various tumor types. 4.2 Hyperparameter Settings HQCM-EBTC was trained for 40 epochs using AdamW (learning rate 0.01, weight decay 1 × 10−4 ). Gradient clipping (max norm 1.0) and Redu…

**Figure 5.** Figure 5: Combined data distribution in the integrated brain tumor dataset. The left panel shows [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]

**Figure 7.** Figure 7: Training and validation loss (left) and accuracy (right) over epochs. The solid lines [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

**Figure 8.** Figure 8: Comparison of t-SNE projections for the Quantum Processing Layer and the Fully Connected Layer. The visualization illustrates the two-dimensional embeddings of test samples, with colors indicating distinct classes. A perplexity of 30 and a fixed random initialization s…

**Figure 9.** Figure 9: Comparison of confusion matrices for HQCM-EBTC and its Classical Counterpart, illus [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

**Figure 10.** Figure 10: Comparison of attention maps generated by the quantum-enhanced model (HQCM [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 37 canonical work pages

[1]

Robust clinical applica- ble cnn and u-net based algorithm for mri classification and segmentation for brain tu- mor

Atika Akter, Nazeela Nosheen, Sabbir Ahmed, Mariom Hossain, Mohammad Abu Yousuf, Mohammad Ali Abdullah Al- moyad, Khondokar Fida Hasan, and Mo- hammad Ali Moni. Robust clinical applica- ble cnn and u-net based algorithm for mri classification and segmentation for brain tu- mor. Expert Systems with Applications , 238:122347, 2024

work page 2024
[2]

Mri brain tumor medical images analysis using deep learning techniques: a systematic review

Sabaa Ahmed Yahya Al-Galal, Imad Fakhri Taha Alshaikhli, and MM Abdul- razzaq. Mri brain tumor medical images analysis using deep learning techniques: a systematic review. Health and Technology, 11(2):267–282, 2021

work page 2021
[3]

Key statistics for brain and spinal cord tumors, 2021

American Cancer Society. Key statistics for brain and spinal cord tumors, 2021. Ac- cessed: 2025-02-01

work page 2021
[4]

A new model for brain tumor detection using en- semble transfer learning and quantum varia- tional classifier

Javeria Amin, Muhammad Almas Anjum, Muhammad Sharif, Saima Jabeen, Seife- dine Kadry, and Pablo Moreno Ger. A new model for brain tumor detection using en- semble transfer learning and quantum varia- tional classifier. Computational intelligence and neuroscience, 2022(1):3236305, 2022

work page 2022
[5]

Amin et al

R. Amin et al. A hybrid quantum-classical framework for lesion detection and seg- mentation. IEEE Transactions on Medical Imaging, 42(5):1123–1135, 2023

work page 2023
[6]

Ayadi, W

W. Ayadi, W. Elhamzi, I. Charfi, and M. Atri. Deep cnn for brain tumor classifica- tion. Neural Processing Letters, 53:671–700, 2021

work page 2021
[7]

Loss functions in the era of semantic segmentation: A survey and out- look, 2023

Reza Azad, Moein Heidary, Kadir Yil- maz, Michael H¨ uttemann, Sanaz Karimi- jafarbigloo, Yuli Wu, Anke Schmeink, and Dorit Merhof. Loss functions in the era of semantic segmentation: A survey and out- look, 2023

work page 2023
[8]

Sohaib Alam, Guillermo Alonso-Linaje, B

Ville Bergholm, Josh Izaac, Maria Schuld, Christian Gogolin, Shahnawaz Ahmed, Vishnu Ajith, M. Sohaib Alam, Guillermo Alonso-Linaje, B. AkashNarayanan, Ali Asadi, Juan Miguel Arrazola, Utkarsh Azad, Sam Banning, Carsten Blank, Thomas R Bromley, Benjamin A. Cordier, Jack Ceroni, Alain Delgado, Olivia Di Mat- teo, Amintor Dusko, Tanya Garg, Diego Guala, Ant...

work page 2022

Show all 40 references

[9]

Quantum machine learn- ing

Jacob Biamonte, Peter Wittek, Nicola Pan- cotti, Patrick Rebentrost, Nathan Wiebe, 19 HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification and Seth Lloyd. Quantum machine learn- ing. Nature, 549(7671):195–202, 2017

2017
[10]

I. Chen, F. D. Johansson, and D. Son- tag. Can ai fairness be improved through explainability? Proceedings of the AAAI Conference on Artificial Intelligence , 34(4):5074–5081, 2020

2020
[11]

Brain mri tumour classification using quan- tum classical convolutional neural net ar- chitecture

Rudrajit Choudhuri and Amiya Halder. Brain mri tumour classification using quan- tum classical convolutional neural net ar- chitecture. Neural Computing and Applica- tions, 35(6):4467–4478, 2023

2023
[12]

Quantum convolutional neural net- works

Iris Cong, Soonwon Choi, and Mikhail D Lukin. Quantum convolutional neural net- works. Nature Physics , 15(12):1273–1278, 2019

2019
[13]

Shul- man

Maurizio Corbetta and Gordon L. Shul- man. Control of goal-directed and stimulus- driven attention in the brain. Nature Re- views Neuroscience , 3(3):201–215, March 2002

2002
[14]

Proposal for a reg- ulation laying down harmonized rules on artificial intelligence (artificial intelligence act), 2021

European Commission. Proposal for a reg- ulation laying down harmonized rules on artificial intelligence (artificial intelligence act), 2021. Official Journal of the European Union

2021
[15]

Clas- sification with quantum neural networks on near term processors

Edward Farhi and Hartmut Neven. Clas- sification with quantum neural networks on near term processors. arXiv preprint arXiv:1802.06002, 2018

2018 arXiv
[16]

Hy- brid deep learning approach for brain tumor classification using efficientnetb0 and novel quantum genetic algorithm

Kerem Gencer and G¨ ulcan Gencer. Hy- brid deep learning approach for brain tumor classification using efficientnetb0 and novel quantum genetic algorithm. PeerJ Com- puter Science, 11:e2556, 2025

2025
[17]

Causability and explainability of artificial intelligence in medicine

Andreas Holzinger, Georg Langs, Helmut Denk, Kurt Zatloukal, and Heimo M¨ uller. Causability and explainability of artificial intelligence in medicine. Wiley Interdisci- plinary Reviews: Data Mining and Knowl- edge Discovery, 9(4):e1312, 2019

2019
[18]

Brain tumor dataset

Cheng Jun. Brain tumor dataset. Figshare, 10:m9, 2017

2017
[19]

Brain tumor diagnosis us- ing quantum convolutional neural networks

Muhammad Al-Zafar Khan, Nouhaila In- nan, Abdullah Al Omar Galib, and Mo- hamed Bennai. Brain tumor diagnosis us- ing quantum convolutional neural networks. arXiv preprint arXiv:2401.15804 , 2024

2024 arXiv
[20]

R. Kolb. Fundamentals of Brain Tumors . Academic Press, 1985

1985
[21]

Brain tumour classification us- ing quantum support vector machine learn- ing algorithm

Tarun Kumar, Dilip Kumar, and Gurmo- han Singh. Brain tumour classification us- ing quantum support vector machine learn- ing algorithm. IETE Journal of Research , 70(5):4815–4828, 2024

2024
[22]

Hugo Larochelle and Geoffrey E. Hinton. Learning to combine foveal glimpses with a third-order boltzmann machine. In Ad- vances in Neural Information Processing Systems (NIPS) , 2010

2010
[23]

Litjens, T

G. Litjens, T. Kooi, B. E. Bejnordi, A. A. A. Setio, F. Ciompi, M. Ghafoorian, and B. van Ginneken. A survey on deep learning in medical image analysis. Medical Image Analysis, 42:60–88, 2017

2017
[24]

S. M. Lundberg and S.-I. Lee. A unified ap- proach to interpreting model predictions. In Advances in Neural Information Processing Systems, pages 4765–4774, 2017. 20 HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification

2017
[25]

Cross-entropy loss functions: Theo- retical analysis and applications, 2023

Anqi Mao, Mehryar Mohri, and Yutao Zhong. Cross-entropy loss functions: Theo- retical analysis and applications, 2023

2023
[26]

Ex- plainability and causability for artificial intelligence-supported medical image anal- ysis in the context of the european in vitro diagnostic regulation

H M¨ uller, A Holzinger, M Plass, L Br- cic, C Stumptner, and K Zatloukal. Ex- plainability and causability for artificial intelligence-supported medical image anal- ysis in the context of the european in vitro diagnostic regulation. N Biotechnol, 70:67– 72, Sep 25 2022

2022
[27]

M. L. Rahman, A. W. Reza, and S. I. Shabuj. An internet of things-based auto- matic brain tumor detection system. In- donesian Journal of Electrical Engineer- ing and Computer Science , 25(1):214–222, 2022

2022
[28]

Hybrid quantum graph neural net- work for brain tumor mr image classifica- tion: Hybrid quantum graph neural net- work for brain tumor mr image

SP Rajamohana, Vani Yelamali, Pallavi Soni, Manjunath Vanahalli, and Prabu Prasad. Hybrid quantum graph neural net- work for brain tumor mr image classifica- tion: Hybrid quantum graph neural net- work for brain tumor mr image. Journal of Scientific & Industrial Research (JSIR...

2025
[29]

Brain tumor segmentation of mri images: A comprehensive review on the application of artificial intelligence tools

R Ranjbarzadeh, A Caputo, EB Tirko- laee, S Jafarzadeh Ghoushchi, and M Ben- dechache. Brain tumor segmentation of mri images: A comprehensive review on the application of artificial intelligence tools. Comput Biol Med , 152:106405, Jan 2023

2023
[30]

Quantum support vector ma- chine for big data classification

Patrick Rebentrost, Masoud Mohseni, and Seth Lloyd. Quantum support vector ma- chine for big data classification. Physical Review Letters, 113(13):130503, 2014

2014
[31]

Samek, T

W. Samek, T. Wiegand, and K.-R. M¨ uller. Explainable artificial intelligence: Un- derstanding, visualizing and interpreting deep learning models. arXiv preprint arXiv:1708.08296, 2017

2017 arXiv
[32]

Quantum machine learning in feature hilbert spaces

Maria Schuld and Nathan Killoran. Quantum machine learning in feature hilbert spaces. Physical review letters , 122(4):040504, 2019

2019
[33]

Grad-cam: Visual explanations from deep networks via gradient-based localization

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. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages ...

2017
[34]

Advancing neuroimag- ing with quantum convolutional neural net- works for brain tumor detection

Amrita Ticku, Vaibhav Sangwan, Sanket Balani, Sriti Jha, Sahil Rawat, Anu Rathee, and Deepika Yadav. Advancing neuroimag- ing with quantum convolutional neural net- works for brain tumor detection. Interna- tional Journal of Information Technology , pages 1–8, 2025

2025
[35]

dicom-brain-dataset

TrainingDataPro. dicom-brain-dataset. Ac- cessed: 2025-11-20

2025
[36]

Brain tumor classification from mri scans: a framework of hybrid deep learning model with bayesian optimization and quantum theory-based marine predator algorithm

Muhammad Sami Ullah, Muhammad At- tique Khan, Anum Masood, Olfa Mzoughi, Oumaima Saidani, and Nazik Alturki. Brain tumor classification from mri scans: a framework of hybrid deep learning model with bayesian optimization and quantum theory-based marine predator algorithm. Fron...

2024
[37]

Visualizing data using t-sne

Laurens Van der Maaten and Geoffrey Hin- ton. Visualizing data using t-sne. Journal of machine learning research , 9(11):2579– 2605, 2008. 21 HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification

2008
[38]

Attention is all you need

Ashish Vaswani, Noam Shazeer, Niki Par- mar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems (NeurIPS), volume 30, 2017

2017
[39]

Environmental risk factors of primary brain tumors: A review

Ali´ enor Vienne-Jumeau, Camille Tafani, and D Ricard. Environmental risk factors of primary brain tumors: A review. Revue neurologique, 175(10):664–678, 2019

2019
[40]

Cbam: Con- volutional block attention module, 2018

Sanghyun Woo, Jongchan Park, Joon- Young Lee, and In So Kweon. Cbam: Con- volutional block attention module, 2018. 22 HQCM-EBTC: A Hybrid Quantum-Classical Model for Explainable Brain Tumor Classification Table 3: Quantum and Classical Model Training Configuration Parameter V ...

2018

Pith tools

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

[1] [1]

Robust clinical applica- ble cnn and u-net based algorithm for mri classification and segmentation for brain tu- mor

Atika Akter, Nazeela Nosheen, Sabbir Ahmed, Mariom Hossain, Mohammad Abu Yousuf, Mohammad Ali Abdullah Al- moyad, Khondokar Fida Hasan, and Mo- hammad Ali Moni. Robust clinical applica- ble cnn and u-net based algorithm for mri classification and segmentation for brain tu- mor. Expert Systems with Applications , 238:122347, 2024

work page 2024

[2] [2]

Mri brain tumor medical images analysis using deep learning techniques: a systematic review

Sabaa Ahmed Yahya Al-Galal, Imad Fakhri Taha Alshaikhli, and MM Abdul- razzaq. Mri brain tumor medical images analysis using deep learning techniques: a systematic review. Health and Technology, 11(2):267–282, 2021

work page 2021

[3] [3]

Key statistics for brain and spinal cord tumors, 2021

American Cancer Society. Key statistics for brain and spinal cord tumors, 2021. Ac- cessed: 2025-02-01

work page 2021

[4] [4]

A new model for brain tumor detection using en- semble transfer learning and quantum varia- tional classifier

Javeria Amin, Muhammad Almas Anjum, Muhammad Sharif, Saima Jabeen, Seife- dine Kadry, and Pablo Moreno Ger. A new model for brain tumor detection using en- semble transfer learning and quantum varia- tional classifier. Computational intelligence and neuroscience, 2022(1):3236305, 2022

work page 2022

[5] [5]

Amin et al

R. Amin et al. A hybrid quantum-classical framework for lesion detection and seg- mentation. IEEE Transactions on Medical Imaging, 42(5):1123–1135, 2023

work page 2023

[6] [6]

Ayadi, W

W. Ayadi, W. Elhamzi, I. Charfi, and M. Atri. Deep cnn for brain tumor classifica- tion. Neural Processing Letters, 53:671–700, 2021

work page 2021

[7] [7]

Loss functions in the era of semantic segmentation: A survey and out- look, 2023

Reza Azad, Moein Heidary, Kadir Yil- maz, Michael H¨ uttemann, Sanaz Karimi- jafarbigloo, Yuli Wu, Anke Schmeink, and Dorit Merhof. Loss functions in the era of semantic segmentation: A survey and out- look, 2023

work page 2023

[8] [8]

Sohaib Alam, Guillermo Alonso-Linaje, B

Ville Bergholm, Josh Izaac, Maria Schuld, Christian Gogolin, Shahnawaz Ahmed, Vishnu Ajith, M. Sohaib Alam, Guillermo Alonso-Linaje, B. AkashNarayanan, Ali Asadi, Juan Miguel Arrazola, Utkarsh Azad, Sam Banning, Carsten Blank, Thomas R Bromley, Benjamin A. Cordier, Jack Ceroni, Alain Delgado, Olivia Di Mat- teo, Amintor Dusko, Tanya Garg, Diego Guala, Ant...

work page 2022