List of Papers Browse by Subject Areas Author List
Abstract
The increasing number of CT scan examinations has lead to the need of developing automated tools to support radiologists managing their growing workload. Multi-label abnormality analysis from 3D chest CT scans, including abnormality classification, segmentation and report generation, is therefore a key capability for clinical decision support. However, such tasks remain highly challenging due to the high-dimensional structure of the data, and the wide variety of abnormalities to detect. Existing approaches typically address abnormality classification, segmentation, or report generation in isolation, despite their strong semantic and clinical interdependence. In clinical workflows, radiologists identify abnormalities by simultaneously localizing regions of interest and synthesizing descriptive findings. In light of this, we propose UniCT, the first end-to-end unified framework that jointly performs multi-label abnormality classification, segmentation, and report generation from 3D chest CT scans. UniCT explicitly models cross-task interactions through a multi-task fusion mechanism and a segmentation-conditioned feature modulation, enabling shared spatial and semantic reasoning across tasks. Experimental results on two public datasets demonstrate that joint learning yields consistent mutual gains across tasks, achieving competitive performance and providing an effective inductive bias for CT analysis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1178_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://theodpzz.github.io/projects/unict/
Link to the Dataset(s)
https://huggingface.co/datasets/rajpurkarlab/ReXGroundingCT
https://huggingface.co/datasets/ibrahimhamamci/CT-RATE
https://zenodo.org/records/6406114
BibTex
@InProceedings{DiThe_UniCT_MICCAI2026,
author = { Di Piazza, Theo AND Lazarus, Carole AND Nempont, Olivier AND Boussel, Loic},
title = { { UniCT: A Unified Joint Multi-Task Framework for 3D Chest CT Abnormality Analysis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16886},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The submitted work proposes an end-to-end unified framework for 3D CT analysis that jointly performs abnormality classification, segmentation, and report generation. It uses cross-task interaction via multi-task supervision and feature fusion to produce complementary clinical outputs.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Clear illustrations and well-presented equations that assist understanding
- Ablation study including quantitative analysis
- Implementation details are reasonably well described
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Performance improvements over baselines are marginal
- Contributions are vaguely stated and lack clear novelty
- Noticeable redundancy and structural issues in writing
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
Detailed Comments: 1.Spatial Interpretability in Classification Models: While classification models are often considered limited in spatial interpretability, they can still preserve meaningful spatial cues through intermediate feature representations. The paper does not sufficiently explore or demonstrate how the proposed model improves such spatial interpretability, particularly in clinically relevant contexts (see point 3 and 4).
2.Unclear and Redundant Contributions: The contributions are not well articulated. Points (1) and (2) appear redundant. Point (3) reflects a standard evaluation requirement, and point (4) is not formulated in a way that can be assessed as a concrete contribution.
3.Limited Performance Improvement: The reported improvements are marginal. For example, Table 1 shows very small gains in validation F1 and AUC scores (approximately 0.003–0.005) on RAD-ChestCT. Similarly, Table 2 demonstrates only minor improvements across tasks (+1% in AUC/DSC and ~0.1% in clinical report generation). These limited gains raise concerns about the effectiveness of the proposed approach and suggest that the problem formulation or model design may require further refinement. A stronger justification and analysis of these results are needed.
4.Qualitative Evaluation: The qualitative results are insufficient to support the claims. Presenting a single selected example does not provide convincing evidence of robustness or clinical relevance. A more comprehensive evaluation, ideally including expert (clinician) assessment or multiple representative cases, is encouraged.
5.Image–text alignment: the model may not learn a precise mapping between visual features and their corresponding textual concepts. It could rely on coarse or global correlations, which leads to weak classification performance. This needs to be addressed.
6.Writing Quality and Structure: The paper contains some redundancy, and frequent use of bold text in place of proper subsections. The overall structure would benefit from clearer organization and more concise writing.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Paper aligned with clinical workflows and is generally clear with adequate experimental coverage; however, it lacks sufficient evidence of clinical significance (especially in qualitative evaluation) and robustness, as some quantitative metrics show only marginal improvements.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The paper proposes UniCT, a unified end-to-end multi-task framework for 3D chest CT abnormality analysis that jointly performs multi-label abnormality classification, abnormality segmentation, and report generation within a single model. Its main contribution is the integration of these three clinically related tasks into one supervised framework with explicit cross-task interaction through a Multi-Task Fusion (MTF) module and Segmentation-Conditioned Feature (SCF) modulation, allowing shared spatial and semantic information to improve all tasks. The paper further shows, through experiments on ReXGroundingCT and external classification evaluation on RAD-ChestCT, that this unified design can provide consistent mutual gains over both task-specific baselines and a matched multi-task baseline without the interaction modules.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper addresses an important and clinically meaningful problem by jointly modeling abnormality classification, segmentation, and report generation from 3D chest CT, which is better aligned with radiology workflow than treating these tasks separately. 2.The main contribution, namely the unified end-to-end UniCT framework with explicit cross-task interaction through MTF and SCF, is interesting because it allows shared spatial and semantic information to improve all tasks. 3.The experimental design is strong: the paper includes task-specific baselines, a matched multi-task baseline, component and task-removal ablations, external classification evaluation on RAD-ChestCT, and a complementarity study with frozen COLIPRI features. 4.The ablation results provide good evidence that the three tasks help each other, which supports the central claim that unified learning is beneficial for 3D chest CT abnormality analysis.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The paper’s novelty is mainly at the unified framework level rather than at the module level: most components are standard, and prior work already cited in the manuscript has addressed related subsets of tasks or broader multi-task settings, such as CT2Rep [17], CT-AGRG [11], Gu et al. [15], MTMed3D [23], Zhang et al. [38], and Tanida et al. [32]. 2.The cross-task interaction is relatively coarse, since the main fusion happens through only three global task embeddings and a global segmentation-conditioned modulation, rather than through richer region-token grounding. 3.While classification results are convincing, the absolute segmentation and report-generation performance remains low, and the paper itself states that performance across all tasks is still below clinically acceptable thresholds. 4.External validation is available only for classification, not for the full unified setting, because RAD-ChestCT lacks reports and segmentation masks. 5.The qualitative analysis is limited to selected successful cases and uses manually drawn reference boxes rather than full ground-truth mask overlays, which weakens the evidence for segmentation quality.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I believe the paper makes a useful contribution and is above the acceptance threshold because the formulation is meaningful, the methodology is coherent, and the empirical evaluation supports the central idea. However, the limitations in module-level novelty and the still modest performance on some tasks keep it in the Weak Accept range rather than a stronger recommendation
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
This paper proposes to combine three tasks (including abnormality classification, segmentation, and report generation) into one unified framework by developing the UniCT model. The key hypothesis is that joint multi-task learning can improve individual tasks.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
-
The idea that joint multi-task learning benefits individual task is well motived and intuitive.
-
The experiments indeed showed performance gains for individual tasks when joint multi-task learning technique is used.
-
The paper is well written and organized.
-
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The performance of each individual task after joint multi-task learning does not look very impressive though improvements over baseline were observed, e.g., AUROC 0.743 for classification and DICE 0.113 for segmentation. It’s hard to say how meaningful a dice 0.113 is for segmentation.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Novel and intuitive idea whose effectiveness was validated with sufficient experimental results.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We sincerely thank the reviewers and the meta-reviewer for their time, careful evaluation, and valuable feedback during the review process.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
This paper proposes UniCT, a unified multi-task framework for 3D chest CT analysis that jointly performs classification, segmentation, and report generation through cross-task interaction. Overall, the paper received consistent Weak Accept–Accept recommendations.
The main strengths are:
- A well-motivated unified framework that aligns with clinical workflows and demonstrates cross-task complementarity;
- Coherent methodology and solid experimental design, including ablations and external validation for classification;
- Empirical evidence that multi-task learning provides consistent, if modest, improvements.
The main weaknesses include:
- Limited novelty at the module level, with most components being standard;
- Marginal performance gains and relatively low absolute performance on some tasks (especially segmentation and report generation);
- Some clarity and writing issues, and limited qualitative validation.
Despite modest improvements and incremental technical novelty, the paper presents a well-executed and clinically meaningful unified framework, with consistent experimental support for its central idea. The contribution is considered sufficient for acceptance, particularly given its relevance and potential to stimulate further work on integrated multi-task learning in medical imaging.