REVIEW 3 major objections 5 minor 53 references
CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding
T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read CARVE compresses 3D medical volumes to about one fifth of their visual tokens before the LLM, retaining 87% of full-token report-generation quality and near-full VQA accuracy.
desk verdict CARVE is a legitimately useful, well-evaluated training-free token compression method for 3D medical MLLMs; the empirical case is strong, though the feature-space proxy for token importance is under-validated and hyperparameters are unreported. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the 2.5D budget-allocation pipeline, built from three scoring functions and two selection loops. Inter-slice change scores $g_t = 1 - \cos(\bar{f}_t, \bar{f}_{t+1})$ define adaptive depth windows; token-level residuals $\tilde{r}_j$, computed against aligned neighbours on adjacent slices with the slice median subtracted, isolate local deviations; the allocation splits the budget into $K_{\text{inter}} = \lfloor \rho K \rfloor$ for global retrieval and the rest for anchors, distributed over windows by a softmax over normalised window statistics. Intra-slice anchor selection uses a best-first quadtree whose refinement priority $\pi(B)$ combines block heterogeneity with the averaged last-layer self-attention saliency $a_j$, placing anchors on representative slices. Global retrieval then ranks all non-anchor tokens by the volume-normalised residual, keeps the strongest quartile as candidates, and applies 3D non-maximum suppression with a spacing-aware Chebyshev radius in physical coordinates to enforce dispersion before the retrieved tokens are kept fixed. Finally, remaining eligible tokens are folded into their closest same-window anchor with a cosine-weighted mean, and the compressed sequence retains its 3D position indices.
What would settle it
Take a random sample of AMOS-MM volumes at the r=0.2 budget and compare per-query answers from the full model against the model with CARVE's dropped tokens manually restored one at a time; if restoring tokens that CARVE assigns low residual scores changes answers on cases where all compared baselines are correct, the proxy scores are missing task-critical evidence. Alternatively, annotate the ground-truth finding regions on AMOS-MM and measure whether CARVE's retrieved tokens concentrate on those regions; if they do not on cross-slice relation questions, the residual objective is not tracking the evidence the benchmark rewards.
Extended reading notes
Core claim
The central claim is that the redundant visual sequence produced by slice-wise encoding of a 3D volume is best compressed not by a flat importance ranking or by dropping whole slices, but by an anisotropic allocation that distinguishes two roles: a spatial scaffold of anchors that summarise recurring anatomy, and a dispersed set of retrieved tokens that capture locally changing findings on any slice. Because the depth axis repeatedly re-images aligned anatomy with sparse embedded changes, token importance is not isotropic in the volume, and a single flat ranking can waste budget on correlated tokens while missing local deviations. CARVE operationalises this by splitting the budget across depth windows and in-plane positions, and by keeping every slice eligible for retrieval rather than committing to a subset of planes. The paper reports that this allocation, rather than a larger budget, drives the quality-efficiency frontier: at a 20% target keep ratio CARVE uses fewer retained tokens than any baseline while scoring highest on every AMOS-MM report metric and near-full retention on VQA tasks.
Load-bearing premise
The method assumes that its three proxy scores—mean-pooled slice cosine similarity, per-token cross-slice residual, and last-layer self-attention—identify the evidence the projector and LLM actually need, even though the paper does not validate whether dropped tokens would have changed a prediction.
Editorial extensions
If this is right
- At a 20% target keep ratio, CARVE leads every compared compression method on all four AMOS-MM report-generation metrics, with an aggregate retention of 87.07% versus 80.89% for the strongest baseline and a range of 66.76-80.89% for the other baselines.
- On 3D-RAD and M3D-VQA, CARVE preserves 96.97% and 99.08% of full-token aggregate performance at realized keep ratios of 19.7% and 19.2%, respectively, and is first on seven of the eight open-ended scores across the two benchmarks.
- The gains transfer across backbones: CARVE achieves the highest aggregate retention on Hulu-Med-4B, Qwen3-VL-4B, and Qwen3-VL-8B, with the best mean rank (2.42) across three backbones and six evaluation tracks.
- The ablation shows both selection branches are necessary: running intra-slice anchors alone drops 1.69 ROUGE-L points on AMOS-MM report generation, inter-slice retrieval alone drops 1.92 points, and dropping folding entirely costs 3.40 points.
- Quality improves with budget for all methods, but the spread narrows as the budget grows: at the most binding ratio CARVE leads by 0.48 ACC and 1.76 ROUGE-L, while near the 0.4 ratio all methods approach full-token quality.
Reading between the lines
- If the paper's proxy scores were validated against the projector and LLM's actual evidence needs, a natural next test is whether the budget split $\rho$ should adapt per question type; the paper uses a fixed $\rho=0.25$, but report generation and closed-ended VQA may require different retrieval shares.
- The scaling finding that in-plane resolution beats slice count at matched token budgets is stated for a specific 2D-encoder stack; a testable extension is whether the same ordering holds for volumetric encoders or lower-resolution whole-volume inputs, which would change acquisition and compression guidance.
- CARVE's saliency signal is the frozen 2D encoder's last-layer attention; replacing it with a task-conditioned signal, such as the language model's cross-attention, is a plausible way to close the residual gap to full tokens, but would require the training-free property to be re-examined.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces CARVE, a training-free token compression framework for slice-based 3D medical vision-language models. It profiles cross-slice feature change, partitions the volume into adaptive windows, splits a target token budget between intra-slice anchors and global inter-slice retrieval, and folds remaining tokens into nearby anchors before the frozen projector and LLM. The authors present scaling analyses suggesting diminishing returns for larger token budgets, and evaluate CARVE on AMOS-MM report generation and VQA, 3D-RAD, and M3D-VQA with Hulu-Med-7B, plus transfer to Hulu-Med-4B and Qwen3-VL. At roughly 20% retained tokens, CARVE reports the highest relative retention among compressed baselines on AMOS-MM report generation (Rel. 87.07 vs. the best baseline 80.89) and preserves 96.97-99.08% of full-token performance on the three VQA benchmarks, supported by a budget sweep and ablations that isolate window allocation, selection roles, and folding.
Significance. If the reported results hold, CARVE is a practically useful drop-in compressor for slice-based 3D medical MLLMs: it is training-free, operates at the post-encoder pre-projector interface, and achieves substantial latency/memory reductions with modest aggregate quality loss. The paper's strengths are its breadth of evaluation (three benchmarks, multiple backbones, a budget sweep, a transfer study, and an ablation design that isolates individual factors) and the clarity of its comparative claims. The main risk is that the method's core scoring proxy—encoder cosine similarity and self-attention—is never directly validated against task relevance, and the aggregate relative-retention metric may mask failures on rare localized findings. These issues are addressable, but they are load-bearing for the central quality-efficiency claim.
major comments (3)
- [Method, Eqs. (1)-(2), (6)] The entire allocation is driven by scoring rules computed in the frozen 2D encoder's feature space: slice change by mean-pooled cosine similarity (Eq. 1), localized deviation by cosine residual to aligned neighbors (Eq. 2), and anchor usefulness by last-layer self-attention (Eq. 6). The paper never validates that these scores reflect what the projector and LLM need for the target tasks. I ask for a direct diagnostic, for example measuring how often dropped tokens are actually required for a correct answer (by comparing predictions when high-scoring versus low-scoring tokens are removed), or stratifying the evaluation by finding size or localization. The single qualitative example in Fig. 7 is suggestive but cannot carry this load. Without such validation, the reported aggregate 'Rel.' scores cannot rule out systematic loss of rare, task-critical evidence.
- [§4.1, Tables 1-2, Fig. 5] All results appear to come from a single run per condition; no error bars, confidence intervals, or significance tests are reported. Some of the decisive margins are small (e.g., M3D-VQA ACC: CARVE 76.97 vs. MMTok 77.01; M3D-VQA BERT: CARVE 56.60 vs. FastVID 56.61), so the claim that CARVE 'leads all compression baselines' is not statistically established. In addition, the method introduces several hyperparameters (rho, K_slice, tau, r_nms, kappa, tau_m) that are fixed to single values, with only the budget r swept. A sensitivity analysis over at least rho and K_slice is needed to confirm that the reported quality-efficiency frontier is not an artifact of a particular hyperparameter setting.
- [§4.2, Fig. 4, Table 3] The evaluation never isolates the failure mode the method is designed to prevent: localized evidence that appears on only a few slices. The task-type breakdown in Fig. 4 reports aggregate accuracy per category, and the ablation table reports aggregate ACC and ROUGE-L, so a method that preserves coarse anatomy while dropping small findings would not be detected. I request a per-case or per-finding recall measure (e.g., accuracy on cases with small lesions, or retrieval recall against annotated regions) in addition to the aggregate metrics. This is especially important because the paper's stated motivation is maintaining a recall path for localized cross-slice evidence.
minor comments (5)
- [Eq. (2)] The definition of N_z(j) as 'available tokens at the same in-plane coordinate on slices z_j +/- 1' should specify how ties and missing coordinates are handled, especially if slice grids are not perfectly aligned.
- [Fig. 1] The figure repeats '#Tok: 2073' for several methods and uses a 'Better' arrow without a legend, making it hard to identify which symbol corresponds to which method; please add a legend and unique labels.
- [Table 3 caption] The caption uses 'full method' to refer to the complete CARVE model, while Tables 1-2 use 'Full' for the uncompressed reference; please rename one of these to avoid confusion.
- [Fig. 2] The scaling analysis would benefit from a precise statement of the slice counts and in-plane resolutions used, since the claim that in-plane resolution beats extra slices is central to the motivation.
- [Eq. (9)] There is a typo: 'The scores j remains fixed' should read 'The score s_j remains fixed.'
Circularity Check
No circularity: CARVE's quality claims are empirical and externally benchmarked, not derived from its scoring equations.
full rationale
CARVE is an empirical method paper. Its 'derivation' consists of algorithmic definitions (Eqs. 1-10) that select and merge tokens; it never claims to derive performance from those scores. The headline numbers (6.2-point gain, 98.1% retention) are measured on external benchmarks (AMOS-MM, 3D-RAD, M3D-VQA) relative to an uncompressed Full model, with fixed hyperparameters and no fitted parameter that is later renamed a prediction. The only potentially circular elements would be if the scoring equations were equivalent to the evaluation metrics by construction, which they are not: cosine similarity and attention scores (Eqs. 1, 2, 6) are not the report or VQA metrics. The paper's untested assumption that these proxies reflect task-relevant importance is a robustness or correctness risk, not circular reasoning, since it is empirically falsifiable and the paper provides direct comparisons to external baselines. There are minor self-citations (Yi et al. 2026; Li et al. 2026b) in related work, but they merely motivate the design and are not load-bearing for the reported results. Hence no circular step can be exhibited, and the appropriate score is 1 (essentially no circularity, with a mild tuning or proxy caveat).
Assumptions & free parameters
free parameters (8)
- target keep ratio r =
0.2 (also 0.4, 0.6, 0.8 in sweep)
- rho (inter-slice budget fraction) =
0.25
- K_slice (anchor density per retained slice) =
64
- tau (window boundary sensitivity, Eq. 3) =
not reported
- r_nms (3D-NMS suppression radius, Eq. 8) =
not reported
- kappa (folding neighborhood radius, Eq. 10) =
not reported
- tau_m (folding temperature, Eq. 10) =
not reported
- heuristic constants in Eqs. 5 and 7 =
0.5 and 0.25
assumptions (4)
- domain assumption Adjacent slices are aligned enough that tokens at the same in-plane coordinate on slices z-1 and z+1 are comparable.
- domain assumption Cosine similarity in the frozen 2D encoder feature space is a valid proxy for cross-slice redundancy and information loss.
- domain assumption The last-layer self-attention of the 2D encoder indicates which tokens are important to the LLM.
- standard math The largest-remainder integerization and softmax allocation in Eq. 5 preserve the intended budget and ordering.
Cite this review
Pith. "Pith review of CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding." pith.science (2026). https://pith.science/paper/CSGYX3OD
@misc{pith2026260804515,
author = {Pith},
title = {Pith review of: CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding},
year = {2026},
howpublished = {\url{https://pith.science/paper/CSGYX3OD}},
note = {Machine review of arXiv:2608.04515}
}
read the original abstract
Slice-based MLLMs leverage mature 2D encoders by representing 3D volumes as sequences of 2D slices. However, this slice-wise formulation produces thousands of visual tokens that burden the LLM backbone, many of which capture overlapping visual evidence across adjacent slices. To understand how effectively a growing visual token budget improves performance, we perform scaling analyses on two 3D medical VQA benchmarks and find diminishing returns: cost keeps rising while accuracy saturates, and improving in-plane resolution is more effective than adding slices at comparable budgets. The budget should therefore be allocated more selectively rather than simply enlarged, yet most token compression methods are designed for 2D images or videos, where redundancy arises from spatial layout or temporal motion rather than from near-duplicate content along the depth axis. We present CARVE, a training-free framework that compresses visual tokens prior to LLM inference and casts token reduction as budget-constrained 2.5D allocation. CARVE partitions the depth axis into coherent windows and allocates tokens non-uniformly according to normalized cross-slice evidence. Under a shared budget, CARVE builds spatial anchors on representative slices and retrieves locally varying evidence from the full volume, then merges remaining eligible tokens into nearby anchors within each window. Removing roughly 80% of the visual tokens on Hulu-Med-7B, CARVE leads all compression baselines on every AMOS-MM report-generation metric, with 6.2 points higher retention of full-token quality than the strongest baseline, and preserves 98.1% of full-token performance across three VQA benchmarks.
Figures
Figures from the paper (4 more)
Reference graph
Works this paper leans on
-
[1]
Proceedings of the 21st annual international ACM SIGIR conference on Research and development in information retrieval , pages=
The use of MMR, diversity-based reranking for reordering documents and producing summaries , author=. Proceedings of the 21st annual international ACM SIGIR conference on Research and development in information retrieval , pages=
-
[2]
arXiv preprint arXiv:2510.08668 , year=
Hulu-med: A transparent generalist model towards holistic medical vision-language understanding , author=. arXiv preprint arXiv:2510.08668 , year=
-
[3]
arXiv preprint arXiv:2603.25155 , year=
Photon: Speedup volume understanding with efficient multimodal large language models , author=. arXiv preprint arXiv:2603.25155 , year=
-
[4]
MedPruner: Training-Free Hierarchical Token Pruning for Efficient 3D Medical Image Understanding in Vision-Language Models , author=. arXiv preprint arXiv:2603.11625 , year=
-
[5]
Advances in Neural Information Processing Systems , volume=
3d-rad: A comprehensive 3d radiology med-vqa dataset with multi-temporal analysis and diverse diagnostic tasks , author=. Advances in Neural Information Processing Systems , volume=
-
[6]
Proceedings of the IEEE/CVF international conference on computer vision , pages=
Sigmoid loss for language image pre-training , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=
-
[7]
arXiv preprint arXiv:2409.12191 , year=
Qwen2-vl: Enhancing vision-language model's perception of the world at any resolution , author=. arXiv preprint arXiv:2409.12191 , year=
-
[8]
arXiv preprint arXiv:2411.14164 , year=
Fopru: Focal pruning for efficient large vision-language models , author=. arXiv preprint arXiv:2411.14164 , year=
Show all 53 references
-
[9]
Proceedings of the Computer Vision and Pattern Recognition Conference , pages=
Divprune: Diversity-based visual token pruning for large multimodal models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=
-
[10]
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
Visionzip: Longer is better but not necessary in vision language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
-
[11]
Advances in Neural Information Processing Systems , volume=
Scope: Saliency-coverage oriented token pruning for efficient multimodel llms , author=. Advances in Neural Information Processing Systems , volume=
-
[12]
Advances in Neural Information Processing Systems , volume=
Beyond attention or similarity: Maximizing conditional diversity for token pruning in mllms , author=. Advances in Neural Information Processing Systems , volume=
-
[13]
European Conference on Computer Vision , pages=
An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models , author=. European Conference on Computer Vision , pages=. 2024 , organization=
2024
-
[14]
arXiv preprint arXiv:2508.18264 , year=
Mmtok: Multimodal coverage maximization for efficient inference of vlms , author=. arXiv preprint arXiv:2508.18264 , year=
-
[15]
arXiv preprint arXiv:2603.29437 , year=
SeGPruner: Semantic-Geometric Visual Token Pruner for 3D Question Answering , author=. arXiv preprint arXiv:2603.29437 , year=
-
[16]
arXiv preprint arXiv:2604.18260 , year=
Geometry-Guided 3D Visual Token Pruning for Video-Language Models , author=. arXiv preprint arXiv:2604.18260 , year=
-
[17]
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
Zero-shot 3d question answering via voxel-based dynamic token compression , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
-
[18]
arXiv preprint arXiv:2308.02463 , year=
Towards generalist foundation model for radiology by leveraging web-scale 2D&3D medical data, arXiv , author=. arXiv preprint arXiv:2308.02463 , year=
-
[19]
Advances in Neural Information Processing Systems , volume=
Llava-med: Training a large language-and-vision assistant for biomedicine in one day , author=. Advances in Neural Information Processing Systems , volume=
-
[20]
arXiv preprint arXiv:2508.02324 , year=
Qwen-image technical report , author=. arXiv preprint arXiv:2508.02324 , year=
-
[21]
arXiv preprint arXiv:2507.05201 , year=
Medgemma technical report , author=. arXiv preprint arXiv:2507.05201 , year=
-
[22]
arXiv preprint arXiv:2403.15388 , year=
Llava-prumerge: Adaptive token reduction for efficient large multimodal models , author=. arXiv preprint arXiv:2403.15388 , year=
-
[23]
Advances in Neural Information Processing Systems , volume=
Fastvid: Dynamic density pruning for fast video large language models , author=. Advances in Neural Information Processing Systems , volume=
-
[24]
Findings of the Association for Computational Linguistics: ACL 2025 , pages=
Prunevid: Visual token pruning for efficient video large language models , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=
2025
-
[25]
Proceedings of the AAAI Conference on Artificial Intelligence , volume=
What kind of visual tokens do we need? training-free visual token pruning for multi-modal large language models from the perspective of graph , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=
-
[26]
arXiv preprint arXiv:2404.00578 , year=
M3d: Advancing 3d medical image analysis with multi-modal large language models , author=. arXiv preprint arXiv:2404.00578 , year=
-
[27]
Advances in neural information processing systems , volume=
Amos: A large-scale abdominal multi-organ benchmark for versatile medical image segmentation , author=. Advances in neural information processing systems , volume=
-
[28]
Pattern Recognition Letters , volume=
Efficient adaptive non-maximal suppression algorithms for homogeneous spatial keypoint distribution , author=. Pattern Recognition Letters , volume=. 2018 , publisher=
2018
-
[29]
Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=
Multi-granular spatio-temporal token merging for training-free acceleration of video llms , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=
-
[30]
arXiv preprint arXiv:2505.16964 , year=
Medframeqa: A multi-image medical vqa benchmark for clinical reasoning , author=. arXiv preprint arXiv:2505.16964 , year=
-
[31]
arXiv preprint arXiv:2504.14692 , year=
Omniv-med: Scaling medical vision-language model for universal visual understanding , author=. arXiv preprint arXiv:2504.14692 , year=
-
[32]
Nature Biomedical Engineering , pages=
Generalist foundation models from a multimodal dataset for 3D computed tomography , author=. Nature Biomedical Engineering , pages=. 2026 , publisher=
2026
-
[33]
Proceedings of the AAAI Conference on Artificial Intelligence , volume=
Hiprune: Training-free visual token pruning via hierarchical attention in vision-language models (student abstract) , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=
-
[34]
arXiv preprint arXiv:2210.09461 , year=
Token merging: Your vit but faster , author=. arXiv preprint arXiv:2210.09461 , year=
-
[35]
arXiv preprint arXiv:2202.07800 , year=
Not all patches are what you need: Expediting vision transformers via token reorganizations , author=. arXiv preprint arXiv:2202.07800 , year=
-
[36]
arXiv preprint arXiv:2606.23888 , year=
E-MRL: Cross-view Aligned Evidence-driven Multimodal Reinforcement Learning for Reliable 3D Tumor Analysis , author=. arXiv preprint arXiv:2606.23888 , year=
-
[37]
arXiv preprint arXiv:2606.23494 , year=
Brain-Adapter: A Dual-Stream Vision-Language MIL Framework for Comprehensive 3D CT Diagnosis of Acute Intracranial Pathologies , author=. arXiv preprint arXiv:2606.23494 , year=
-
[38]
International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=
Sali: Short-term alignment and long-term interaction network for colonoscopy video polyp segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=
2024
-
[39]
arXiv preprint arXiv:2501.08458 , year=
Rwkv-unet: Improving unet with long-range cooperation for effective medical image segmentation , author=. arXiv preprint arXiv:2501.08458 , year=
-
[40]
arXiv preprint arXiv:2601.08758 , year=
M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding , author=. arXiv preprint arXiv:2601.08758 , year=
-
[41]
Medical Image Analysis , pages=
Read like a radiologist: efficient vision-language model for 3D medical imaging interpretation , author=. Medical Image Analysis , pages=. 2026 , publisher=
2026
-
[42]
IEEE Journal of Biomedical and Health Informatics , year=
Med3dvlm: An efficient vision-language model for 3d medical image analysis , author=. IEEE Journal of Biomedical and Health Informatics , year=
-
[43]
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
Samix: Reinforcing sam2 with semantic adapter and reference selecting policy for mix-supervised segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
-
[44]
IEEE Transactions on Medical Imaging , year=
Enhancing Knee Disease Diagnosis via Multi-View Graph Representation With Multi-Task Pre-Training , author=. IEEE Transactions on Medical Imaging , year=
-
[45]
Neurocomputing , volume=
REHRSeg: Unleashing the power of self-supervised super-resolution for resource-efficient 3D MRI segmentation , author=. Neurocomputing , volume=. 2025 , publisher=
2025
-
[46]
Research Square , pages=
Merlin: A vision language foundation model for 3d computed tomography , author=. Research Square , pages=
-
[47]
International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=
Ct2rep: Automated radiology report generation for 3d medical imaging , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=
2024
-
[48]
Proceedings of the Computer Vision and Pattern Recognition Conference , pages=
Pact: Pruning and clustering-based token reduction for faster visual language models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=
-
[49]
Nature Communications , volume=
Towards a holistic framework for multimodal LLM in 3D brain CT radiology report generation , author=. Nature Communications , volume=. 2025 , publisher=
2025
-
[50]
arXiv preprint arXiv:2508.21044 , year=
Mmg-vid: Maximizing marginal gains at segment-level and token-level for efficient video llms , author=. arXiv preprint arXiv:2508.21044 , year=
-
[51]
Advances in Neural Information Processing Systems , volume=
Better tokens for better 3d: Advancing vision-language modeling in 3d medical imaging , author=. Advances in Neural Information Processing Systems , volume=
-
[52]
Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=
Adapting vision-language models for 3d ct/mri understanding on pmbb via slice selection and explanation analysis , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=
-
[53]
Scientific Reports , volume=
A data-efficient 3D medical vision-language model using only a 2D encoder , author=. Scientific Reports , volume=. 2026 , publisher=
2026
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.