Pith. sign in

REVIEW 6 major objections 3 minor 30 references

RadiomicsRetrieval: A Customizable Framework for Medical Image Retrieval Using Radiomics Features

T0 review · 6 major / 3 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 Tumor-level contrastive alignment of image, radiomics, and anatomy turns a point-click prompt into flexible 3D tumor retrieval.

desk verdict A solid engineering contribution with a real evaluation gap in the brain APE pretraining; worth reviewing, but the brain location claims need a clearer data-exclusion statement. read the letter →

arxiv 2507.08546 v1 pith:NI2LMLOG submitted 2025-07-11 cs.CV cs.AI

Inye Na , Nejung Rue , Jiwon Chung , Hyunjin Park This is my paper · ORCID
classification cs.CVcs.AI
keywords medicalimageretrievalradiomicscontrastivelearningpromptablesegmentationanatomicalpositionalembedding3Dimagingflexiblequery
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

RadiomicsRetrieval proposes to make 3D medical image retrieval work at the level of a single tumor rather than a whole scan. It aligns an image embedding, obtained from a promptable segmentation model with just a point click, to a radiomics embedding from the same tumor's shape, intensity, and texture descriptors, in a shared contrastive space, and adds anatomical positional context. If this alignment holds, a user can search a large image repository using an image, a location, or any subset of radiomics attributes, without drawing full tumor boundaries at query time. The authors report that this combination improves retrieval specificity on lung CT and brain MRI compared with image-only embeddings and makes location-based queries effective.

What carries the argument

The central machinery is a shared latent space built by tumor-level multi-positive contrastive learning. The image side uses a promptable volumetric segmentation model, SAM-Med3D, with an augmented decoder that emits a tumor image embedding from a point prompt; the radiomics side uses TransTab to encode 72 PyRadiomics features as tokenized feature name-value pairs, so feature subsets from 72 down to 1 remain queryable without retraining. Same-tumor embeddings under varying prompts and random crops form positives, and other tumors form negatives. Anatomical positional embedding, a voxel-based coordinate map of where the tumor sits in the body, is added element-wise into the image path and concatenated into the radiomics path. The InfoNCE-style contrastive loss, alongside segmentation and classification losses, is what forces the radiomics and image views to be compared on tumor identity rather than global image similarity, and the shared space is what lets a query in any modality be compared against the same reference set.

What would settle it

On a held-out set of the same two datasets, query every tumor with a point placed at the tumor periphery instead of the center; if the top-k radiomics correlation or location precision falls to within a few points of the random baseline, the claim that a single point prompt suffices for tumor-specific retrieval is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that uniting handcrafted radiomics descriptors with deep image embeddings at tumor level, through multi-positive contrastive learning, makes 3D medical image retrieval both more specific and more query-flexible than image-only embeddings. Radiomics features encode shape, histogram, and texture of the same tumor that the image path sees, and the contrastive objective pulls every view of one tumor together while pushing different tumors apart. Anatomical positional embedding (APE) supplies global context, which is what allows location-based and APE-only queries to work. On the reported results, the multimodal image-plus-radiomics setting with APE reaches location precision@10 of 0.9145 on BraTS and 0.9000 on NSCLC, while radiomics-with-image queries raise top-10 radiomics correlation from roughly 0.85 to 0.96 on BraTS and from roughly 0.82 to 0.93 on NSCLC. The authors state the outcome as a customizable retrieval system that accepts an image, an anatomical location, or any partial set of radiomics features as a query.

Load-bearing premise

Everything rests on a single point clicked on a scan being enough for the segmentation-backed encoder to isolate the tumor on unseen data; if that click does not localize the tumor, the tumor-specific embedding is wrong and the radiomics and anatomical alignments built on it degrade.

Editorial extensions

If this is right

  • A single point prompt at the tumor center is enough to build the reference database and to query it, so no full tumor segmentation is needed at inference time.
  • Users can query by anatomical location alone; a one-point APE query localizes tumors accurately in both datasets.
  • Users can query with any subset of the 72 radiomics features, from all features down to a single one, without retraining the model.
  • Adding radiomics to the image path raises the radiomics correlation of retrieved neighbors compared with image-only embeddings, while combining Fourier and anatomical positional encodings preserves location specificity alongside local tumor detail.

Reading between the lines

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

  • The paper does not test how prompt count or placement changes retrieval quality: training saw 1-10 random prompts, but evaluation uses a single tumor-center point. A direct extension would sweep prompt locations and report precision, which could be decisive for clinical adoption.
  • The same design should transfer to other solid-tumor modalities if the anatomical embedding is retrained for that body region; the paper reports only lung CT and brain MRI, so such transfer is an open question.
  • Because a one-feature query such as voxel volume already returns distinct retrievals, this setup could automate cohort construction from large imaging archives, a use the paper motivates but does not quantify.
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

6 major / 3 minor

Summary. The paper proposes RadiomicsRetrieval, a 3D tumor-level content-based image retrieval (CBIR) framework that combines promptable SAM-Med3D image embeddings with handcrafted radiomics feature embeddings through multimodal contrastive learning. The framework uses anatomical positional embeddings (APE) to inject global anatomical context into both the image and radiomics paths, and it supports several query modes: tumor image with a single point prompt, radiomics feature subsets, APE-only location queries, and combinations of these. The method is evaluated on BraTS 2023 brain tumor MRI (GLI and MEN) and on NSCLC lung CT, using location-based precision@k and radiomics-feature correlation metrics. The central claims are that radiomics features significantly enhance retrieval specificity, that APE provides global anatomical context essential for location-based searches, and that minimal user prompts suffice for tumor-specific retrieval.

Significance. If the main claims hold, the framework would be a useful practical contribution: it reduces segmentation overhead via point prompts, enables flexible querying over partial radiomics features, and combines classical radiomics descriptors with learned embeddings in a principled contrastive formulation. The internal ablations are consistent with the qualitative conclusion that multimodal contrastive training improves the chosen metrics over the unimodal image-only baseline, and the authors provide code. The paper does not rely on fitted constants in a circular derivation, and the method description is largely reproducible. However, the empirical evidence has load-bearing gaps: the brain APE pretraining data may overlap with the evaluation patients, the brain location metric is only left/right, no existing CBIR baselines are compared, and there are no error bars or significance tests. These gaps currently prevent the paper from fully supporting its generalizable-anatomy and specificity claims.

major comments (6)
  1. [Section 2.3 and Section 3] The brain APE model is pretrained on BraTS MRI data (Section 2.3), while the evaluation split is drawn from BraTS 2023 GLI and MEN (Section 3) with an 8:1:1 patient split. The manuscript never states that the 179 test and 176 validation patients were excluded from the APE pretraining cohort. If there is patient-level overlap, the brain rows of Tables 1 and 2 could reflect identity leakage rather than anatomical generalization, because APE values at prompt coordinates are concatenated into both the image and radiomics paths. Please report the patient-level disjointness explicitly, or retrain the brain APE on a patient-disjoint cohort and re-report the brain results.
  2. [Section 4, Tables 1 and 2] All reported metrics appear to come from a single training run, with no standard deviations, confidence intervals, or significance tests. Several comparisons are numerically close, such as <Image,Radiomics,APE> versus <Image,Radiomics,FPE+APE> in Table 1 (NSCLC P@5: 0.9115 versus 0.9062), so the claim of 'significant enhancement' is not statistically supported. Please report results over multiple seeds with error bars and, where appropriate, pairwise significance tests.
  3. [Section 3 and Table 1] For brain MRI, the location-based evaluation reduces 'same region' to left/right hemisphere, which is a very coarse anatomic distinction. The lung evaluation uses five lobes, but the brain evaluation uses only two regions. This coarse metric is insufficient to support the core claim that APE provides 'global anatomical context essential for location-based searches'; the authors should evaluate location retrieval at lobe-level or finer parcellations for the brain data.
  4. [Section 4.2, Table 2] The radiomics-based evaluation metric measures correlation with the same 72 radiomics features that are used as a training modality in the contrastive objective. The observed gains of multimodal-contrastive over unimodal models are therefore partly an alignment artifact rather than evidence of clinically relevant retrieval specificity. To strengthen the claim, report retrieval performance against independent semantic labels such as tumor type, subtype, or expert relevance judgments, and compare against a simple radiomics-only nearest-neighbor baseline (the Upper Bound row is a useful reference but is not a retrieval model).
  5. [Section 2.1 and Section 4] During training, the model sees 1 to 10 random point prompts and random crops (Section 2.3), while at inference both database construction and querying use a single tumor-center point prompt (Section 4). The paper does not ablate prompt count or prompt placement. If the augmented SAM-Med3D decoder does not reliably localize the tumor from one click on unseen data, the resulting embeddings are not tumor-specific and the reported alignment gains would not transfer. Please report retrieval metrics as a function of prompt count (e.g., 1, 3, 5, 10) and prompt placement (center versus off-center).
  6. [Section 4] The experimental section contains no comparison with existing CBIR methods, despite the introduction positioning the work against 2D and mask-based approaches. Without baselines such as 3D-MIR [1], a standard 3D encoder embedding, a radiomics-only retrieval system, or a text-based retrieval model, the paper does not establish that the proposed framework is competitive with or superior to the current state of the art. Please add at least one or two standard CBIR baselines to contextualize the reported numbers.
minor comments (3)
  1. [Figure 2] The abbreviations 'ADE SCC LCC NOS' appear in the figure without definition; the text uses 'ADC' for adenocarcinoma in Section 3, creating an inconsistency between the figure and the dataset description.
  2. [Tables 1 and 2] The row labels 'Image & 1 tumor point' and '1 point APE' are awkwardly phrased; suggest 'Image with 1 tumor-center point' and 'APE with 1 point' for clarity, and define the region labels (left/right brain, five lung lobes) directly in the table captions.
  3. [Figure 4] Figure 4 shows example retrieved cases, but the histograms and the selection criteria are not described; it would be clearer to include quantitative values for the displayed radiomics features and to state explicitly how the shown examples were chosen.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity; one mild metric-objective alignment and a non-load-bearing self-citation.

  1. fitted input called prediction [Section 2 (Training Details) and Section 4.2 (Table 2)]
    "Table 2 shows correlations of tumor-internal radiomics features between queries and retrievals. Multimodal-contrastive methods substantially outperformed unimodal-contrastive approaches, demonstrating the benefit of combining image and radiomics paths."

    The multimodal advantage in Table 2 is the direct output of the training objective: L_contrast (InfoNCE) was optimized to align image embeddings with radiomics embeddings of the same tumor. The radiomics-based evaluation then measures the average correlation between the query's raw radiomics features and those of the retrieved tumors, i.e., the same cross-modal alignment the model was explicitly fitted to maximize. Thus the claim that 'radiomics features significantly enhance retrieval specificity' partly restates the training target rather than an independent emergent result. Since the test set is held out and the metric uses raw radiomics values rather than the learned embedding cosine, this is a mild consistency check, not a full identity between fitted input and prediction.

full rationale

The paper's core derivation is a contrastive alignment of tumor image embeddings with radiomics embeddings, and the experiments test this alignment on held-out BraTS and NSCLC splits with external radiomics and atlas-based location metrics. No equation reduces a predicted quantity to a fitted input: APE is a pretrained positional embedding from prior work (ref 13) and is used as an input feature, not as a learned predictor of location; the reported location gains reflect adding location coordinates to the query, which is the intended design rather than a circular derivation. The only mild circularity is the radiomics-based evaluation in Table 2, which uses a metric closely aligned with the InfoNCE objective; this inflates the apparent significance of the radiomics contribution but does not invalidate the held-out comparison. Ref 21 is a same-author background citation for radiomics requiring manual segmentation; it is not load-bearing. The BraTS APE pretraining-overlap concern raised by the skeptic is a data-leakage risk, not a circularity: if the APE model saw the test patients, the brain location results would be invalid, but this would be an evaluation flaw rather than a derivation that reduces to its own inputs. Overall, the central claims are self-contained and benchmarked, so the paper is not materially circular.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

No new entities are introduced. The central claims rest on domain assumptions about radiomics features, anatomical coordinate consistency, and cosine-similarity ranking, plus one fitted component (brain APE pretrained on BraTS) whose data provenance is underspecified.

free parameters (1)
  • Brain APE model pretrained on BraTS MRI data = Learned weights (not released)
    Section 2.3 states the APE model was pretrained with BraTS MRI data for brain tumor MRI; the paper does not state whether evaluation subjects were excluded, and location-based retrieval results depend on this fitted model.
assumptions (3)
  • domain assumption PyRadiomics features are a valid and meaningful description of tumor phenotype (shape, intensity histogram, texture).
    The radiomics path and the radiomics-based evaluation metric (average correlation of top-k features, Table 2) assume these 72 features capture clinically relevant tumor properties; standard in the field but not independently validated here.
  • domain assumption The anatomical positional embedding provides a consistent cross-subject coordinate system for lung lobes and brain hemispheres.
    Section 2.3 integrates APE into both paths, and the location evaluation in Table 1 assumes the same anatomical region labels are comparable across subjects.
  • domain assumption Cosine similarity in the learned shared latent space is a valid ranking criterion for retrieval.
    Inference (Section 2.3, Retrieval Inference) retrieves by cosine similarity; this assumes the contrastive space preserves clinically relevant distances.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RadiomicsRetrieval: A Customizable Framework for Medical Image Retrieval Using Radiomics Features." pith.science (2026). https://pith.science/paper/NI2LMLOG

@misc{pith2026250708546,
  author       = {Pith},
  title        = {Pith review of: RadiomicsRetrieval: A Customizable Framework for Medical Image Retrieval Using Radiomics Features},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NI2LMLOG}},
  note         = {Machine review of arXiv:2507.08546}
}
read the original abstract

Medical image retrieval is a valuable field for supporting clinical decision-making, yet current methods primarily support 2D images and require fully annotated queries, limiting clinical flexibility. To address this, we propose RadiomicsRetrieval, a 3D content-based retrieval framework bridging handcrafted radiomics descriptors with deep learning-based embeddings at the tumor level. Unlike existing 2D approaches, RadiomicsRetrieval fully exploits volumetric data to leverage richer spatial context in medical images. We employ a promptable segmentation model (e.g., SAM) to derive tumor-specific image embeddings, which are aligned with radiomics features extracted from the same tumor via contrastive learning. These representations are further enriched by anatomical positional embedding (APE). As a result, RadiomicsRetrieval enables flexible querying based on shape, location, or partial feature sets. Extensive experiments on both lung CT and brain MRI public datasets demonstrate that radiomics features significantly enhance retrieval specificity, while APE provides global anatomical context essential for location-based searches. Notably, our framework requires only minimal user prompts (e.g., a single point), minimizing segmentation overhead and supporting diverse clinical scenarios. The capability to query using either image embeddings or selected radiomics attributes highlights its adaptability, potentially benefiting diagnosis, treatment planning, and research on large-scale medical imaging repositories. Our code is available at https://github.com/nainye/RadiomicsRetrieval.

Figures

Figures reproduced from arXiv: 2507.08546 by the authors.

Figure 1
Figure 1. Illustration of the RadiomicsRetrieval. Users can retrieve similar tumors by providing either an image with minimal point prompts or customized conditions (e.g., anatomical location or partial radiomics features). This diverse setup supports both image-based queries and more general feature-driven searches. While CBIR methods relying on deep learning–based encoders have surpassed those based on handcrafted features,… view at source ↗
Figure 2
Figure 2. (A) Overall architecture of RadiomicsRetrieval, showing parallel paths for tumor image (green), radiomics (blue), and anatomical (purple) embeddings. (B) Detailed view of how the anatomical positional embedding (APE) is integrated with the Fourier-based positional encoding (FPE). ciency, making RadiomicsRetrieval a highly adaptable CBIR solution. Our main contributions are: 1. Hybrid Radiomics–Deep Learning Approach… view at source ↗
Figure 3
Figure 3. Multi-positive contrastive learning. (A) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Retrieved Samples Under Different Query Types. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 19 canonical work pages

  1. [1]

    arXiv preprint arXiv:2311.13752 (2023)

    Abacha, A.B., Santamaria-Pang, A., Lee, H.H., Merkow, J., Cai, Q., Devarakonda, S.T., Islam, A., Gong, J., Lungren, M.P., Lin, T., et al.: 3d-mir: A benchmark and empirical study on 3d medical image retrieval in radiology. arXiv preprint arXiv:2311.13752 (2023)

    arXiv 2023

  2. [2]

    The Cancer Imaging Archive (2014)

    Aerts, H.J., Wee, L., Rios Velazquez, E., Leijenaar, R.T., Parmar, C., Grossmann, P., Carvalho, S., Bussink, J., Monshouwer, R., Haibe-Kains, B., et al.: Data from nsclc-radiomics. The Cancer Imaging Archive (2014)

    work page 2014

  3. [3]

    arXiv preprint arXiv:2107.02314 (2021)

    Baid, U., Ghodasara, S., Mohan, S., Bilello, M., Calabrese, E., Colak, E., Farahani, K., Kalpathy-Cramer, J., Kitamura, F.C., Pati, S., et al.: The rsna-asnr-miccai brats 2021 benchmark on brain tumor segmentation and radiogenomic classifica- tion. arXiv preprint arXiv:2107.02314 (2021)

    arXiv 2021

  4. [4]

    Scientific data 4(1), 1–13 (2017)

    Bakas, S., Akbari, H., Sotiras, A., Bilello, M., Rozycki, M., Kirby, J.S., Freymann, J.B., Farahani, K., Davatzikos, C.: Advancing the cancer genome atlas glioma mri collections with expert segmentation labels and radiomic features. Scientific data 4(1), 1–13 (2017)

    2017

  5. [5]

    Bakr, S., Gevaert, O., Echegaray, S., Ayers, K., Zhou, M., Shafiq, M., Zheng, H., Zhang,W.,Leung,A.,Kadoch,M.,etal.:Datafornsclcradiogenomics.TheCancer Imaging Archive (2017)

    work page 2017

  6. [6]

    European Radiology Experimental9(1), 4 (2025)

    Böttcher, B., van Assen, M., Fari, R., von Knebel Doeberitz, P.L., Kim, E.Y., Berkowitz, E.A., Meinel, F.G., De Cecco, C.N.: Evaluation of a content-based im- ageretrievalsystemforradiologistsinhigh-resolutionctofinterstitiallungdiseases. European Radiology Experimental9(1), 4 (2025)

    work page 2025

  7. [7]

    In: International conference on machine learning

    Chen, T., Kornblith, S., Norouzi, M., Hinton, G.: A simple framework for con- trastive learning of visual representations. In: International conference on machine learning. pp. 1597–1607. PmLR (2020)

    2020

  8. [8]

    In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention

    Chen, Y., Liu, C., Liu, X., Arcucci, R., Xiong, Z.: Bimcv-r: A landmark dataset for 3d ct text-image retrieval. In: International Conference on Medical Image Com- puting and Computer-Assisted Intervention. pp. 124–134. Springer (2024) 10 I. Na and N. Rue et al

    work page 2024

Show all 30 references
  1. [9]

    Radiology302(1), 187–197 (2022)

    Choe, J., Hwang, H.J., Seo, J.B., Lee, S.M., Yun, J., Kim, M.J., Jeong, J., Lee, Y., Jin, K., Park, R., et al.: Content-based image retrieval by using deep learning for interstitial lung disease diagnosis with chest ct. Radiology302(1), 187–197 (2022)

    2022

  2. [10]

    In: Information Processing in Medical Imaging: 16th International Conference, IPMI’99 Visegrád, Hungary, June 28–July 2, 1999 Proceedings 16

    Collins,D.L.,Zijdenbos,A.P.,Baaré,W.F.,Evans,A.C.:Animal+insect:improved cortical structure segmentation. In: Information Processing in Medical Imaging: 16th International Conference, IPMI’99 Visegrád, Hungary, June 28–July 2, 1999 Proceedings 16. pp. 210–223. Springer (1999)

    1999

  3. [11]

    Neuroimage54(1), 313–327 (2011)

    Fonov, V., Evans, A.C., Botteron, K., Almli, C.R., McKinstry, R.C., Collins, D.L., Group, B.D.C., et al.: Unbiased average age-appropriate atlases for pediatric stud- ies. Neuroimage54(1), 313–327 (2011)

    2011

  4. [12]

    Neu- roImage 47, S102 (2009)

    Fonov, V.S., Evans, A.C., McKinstry, R.C., Almli, C.R., Collins, D.: Unbiased nonlinear average age-appropriate brain templates from birth to adulthood. Neu- roImage 47, S102 (2009)

    2009

  5. [13]

    In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion

    Goncharov, M., Samokhin, V., Soboleva, E., Sokolov, R., Shirokikh, B., Belyaev, M., Kurmukov, A., Oseledets, I.: Anatomical positional embeddings. In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion. pp. 68–77. Springer (2024)

    2024

  6. [14]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 4015–4026 (2023)

    2023

  7. [15]

    Medical Image Analysis92, 103060 (2024)

    Kobayashi, K., Gu, L., Hataya, R., Mizuno, T., Miyake, M., Watanabe, H., Taka- hashi, M., Takamizawa, Y., Yoshida, Y., Nakamura, S., et al.: Sketch-based seman- tic retrieval of medical images. Medical Image Analysis92, 103060 (2024)

    2024

  8. [16]

    Journal of digital imaging26, 1025–1039 (2013)

    Kumar, A., Kim, J., Cai, W., Fulham, M., Feng, D.: Content-based medical image retrieval: a survey of applications to multidimensional and multimodality data. Journal of digital imaging26, 1025–1039 (2013)

    2013

  9. [17]

    arXiv preprint arXiv:2305.07642 (2023)

    LaBella, D., Adewole, M., Alonso-Basanta, M., Altes, T., Anwar, S.M., Baid, U., Bergquist, T., Bhalerao, R., Chen, S., Chung, V., et al.: The asnr-miccai brain tu- mor segmentation (brats) challenge 2023: Intracranial meningioma. arXiv preprint arXiv:2305.07642 (2023)

    2023 arXiv

  10. [18]

    The cancer imaging archive (2020)

    Li, P., Wang, S., Li, T., Lu, J., HuangFu, Y., Wang, D.: A large-scale ct and pet/ct dataset for lung cancer diagnosis (lung-pet-ct-dx). The cancer imaging archive (2020)

    2020

  11. [19]

    IEEE transactions on medical imaging 34(10), 1993–2024 (2014)

    Menze, B.H., Jakab, A., Bauer, S., Kalpathy-Cramer, J., Farahani, K., Kirby, J., Burren, Y., Porz, N., Slotboom, J., Wiest, R., et al.: The multimodal brain tumor image segmentation benchmark (brats). IEEE transactions on medical imaging 34(10), 1993–2024 (2014)

    2014

  12. [20]

    International journal of medical informatics73(1), 1–23 (2004)

    Müller, H., Michoux, N., Bandon, D., Geissbuhler, A.: A review of content-based image retrieval systems in medical applications—clinical benefits and future direc- tions. International journal of medical informatics73(1), 1–23 (2004)

    2004

  13. [21]

    Frontiers in Oncology14, 1341228 (2024)

    Na, I., Noh, J.J., Kim, C.K., Lee, J.W., Park, H.: Combined radiomics-clinical model to predict platinum-sensitivity in advanced high-grade serous ovarian carci- noma using multimodal mri. Frontiers in Oncology14, 1341228 (2024)

    2024

  14. [22]

    Information Sciences637, 118938 (2023)

    Öztürk, Ş., Çelik, E., Çukur, T.: Content-based medical image retrieval with op- ponent class adaptive margin loss. Information Sciences637, 118938 (2023)

    2023

  15. [23]

    Neurocomputing266, 8–20 (2017)

    Qayyum, A., Anwar, S.M., Awais, M., Majid, M.: Medical image retrieval using deep convolutional neural network. Neurocomputing266, 8–20 (2017)

    2017

  16. [24]

    Advances in neural informa- tion processing systems33, 7537–7547 (2020)

    Tancik, M., Srinivasan, P., Mildenhall, B., Fridovich-Keil, S., Raghavan, N., Sing- hal, U., Ramamoorthi, R., Barron, J., Ng, R.: Fourier features let networks learn RadiomicsRetrieval 11 high frequency functions in low dimensional domains. Advances in neural informa- tion pro...

    2020

  17. [25]

    Cancer research77(21), e104–e107 (2017)

    Van Griethuysen, J.J., Fedorov, A., Parmar, C., Hosny, A., Aucoin, N., Narayan, V., Beets-Tan, R.G., Fillion-Robin, J.C., Pieper, S., Aerts, H.J.: Computational radiomics system to decode the radiographic phenotype. Cancer research77(21), e104–e107 (2017)

    2017

  18. [26]

    arXiv preprint (2023)

    Wang, H., Guo, S., Ye, J., Deng, Z., Cheng, J., Li, T., Chen, J., Su, Y., Huang, Z., Shen, Y., et al.: Sam-med3d: towards general-purpose segmentation models for volumetric medical images. arXiv preprint (2023)

    2023

  19. [27]

    Advances in Neural Information Processing Systems35, 2902–2915 (2022)

    Wang, Z., Sun, J.: Transtab: Learning transferable tabular transformers across tables. Advances in Neural Information Processing Systems35, 2902–2915 (2022)

    2022

  20. [28]

    Radiology: Artificial Intelligence 5(5) (2023)

    Wasserthal, J., Breit, H.C., Meyer, M.T., Pradella, M., Hinck, D., Sauter, A.W., Heye, T., Boll, D.T., Cyriac, J., Yang, S., et al.: Totalsegmentator: robust segmen- tation of 104 anatomic structures in ct images. Radiology: Artificial Intelligence 5(5) (2023)

    2023

  21. [29]

    The Cancer Imaging Archive (2019)

    Wee, L., Aerts, H., Kalendralis, P., Dekker, A.: Data from nsclc-radiomics- interobserver1. The Cancer Imaging Archive (2019)

    2019

  22. [30]

    The Cancer Imaging Archive (2020)

    Wee, L., Aerts, H., Kalendralis, P., Dekker, A.: Rider lung ct segmentation labels from: Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. The Cancer Imaging Archive (2020)

    2020

Pith tools

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