mlsanityData to 14 Oct 2026

SelfCue: Making a 3D CT Report Generator Say What It Already Knows

arXiv 2609.31788 ↗First posted 25 Sep 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Progress in 3D CT report generation is usually sought in increasingly sophisticated architectures and larger pools of training data. We find instead that a 3D CT report generator already holds what its report leaves out, and loses it when the hidden state becomes tokens. Over the 18 CT-RATE abnormalities, this hidden-to-report surfacing gap is reflected by a drop in macro AUROC from 0.848 in the hidden states to 0.739 in the generated report. We propose SelfCue based on contrastive decoding. It promotes what the hidden state already supports and suppresses what it does not. It raises clinical efficacy F1 to 0.481 and the LLM-judged GREEN score to 0.510. Distilling that behaviour into the weights gives SelfCue-KD, a student that keeps most of the gain, needs nothing extra at inference, and drops into any pipeline already serving the baseline. Code is available at https://github.com/renjie-liang/SelfCue-CT.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 36 of its 49 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

36 of 49 references.

References · 36

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  2. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  3. 2023Visual Instruction Tuningcited by 7,869
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  5. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  6. 2016Sequence-Level Knowledge Distillationcited by 801
  7. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 767
  8. 2022On Distillation of Guided Diffusion Modelscited by 725
  9. 2023Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decodingcited by 688
  10. 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 609
  11. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 327
  12. 2024LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinationscited by 244
24 more
  1. 2024Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomographycited by 199
  2. 2022Learning by Distilling Contextcited by 138
  3. 2024GREEN: Generative Radiology Report Evaluation and Error Notationcited by 128
  4. 2024CT2Rep: Automated Radiology Report Generation for 3D Medical Imagingcited by 96
  5. 2023PromptMRG: Diagnosis-Driven Prompts for Medical Report Generationcited by 95
  6. 2023Stay on topic with Classifier-Free Guidancecited by 91
  7. 2021Knowledge Distillation as Semiparametric Inferencecited by 29
  8. 2025Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imagingcited by 20
  9. 2024Toward Guidance-Free AR Visual Generation via Condition Contrastive Alignmentcited by 14
  10. 2025Boosting Vision Semantic Density with Anatomy Normality Modeling for Medical Vision-language Pre-trainingcited by 14
  11. 2024CT-AGRG: Automated Abnormality-Guided Report Generation from 3D Chest CT Volumescited by 10
  12. 2026OmniCT: Towards a Unified Slice-Volume LVLM for Comprehensive CT Analysiscited by 10
  13. 2026RadAgent: A tool-using AI agent for stepwise interpretation of chest computed tomographycited by 6
  14. 2025CCD: Mitigating Hallucinations in Radiology MLLMs via Clinical Contrastive Decodingcited by 4
  15. 2025MEPNet: Medical Entity-balanced Prompting Network for Brain CT Report Generationcited by 3
  16. 2026Beyond the Embedding Bottleneck: Adaptive Retrieval-Augmented 3D CT Report Generationcited by 3
  17. 2026CWCD: Category-Wise Contrastive Decoding for Structured Medical Report Generationcited by 3
  18. 2026Generating Reports or Repeating Templates? Measuring and Mitigating Template Collapse in 3D CT Report Generationcited by 3
  19. 2026Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimescited by 2
  20. 2026SliceWorld: A Predictive and Controllable World-State Model for CT Report Generationcited by 1
  21. 2026Not All Tokens Matter Equally: Dynamic In-context Vector Distillation with Decisive-Token Supervision for Long-form Medical Report Generationcited by 1
  22. 2026When Do Cheap Probes Predict Expensive Training? Probing 3D-CT Encoders for Text Generationcited by 1
  23. 2026ORCA: ORgan-Centroid Aggregation for Training-Free 3D CT Visual Token Compressioncited by 1
  24. 2026The Knowing-Saying Gap: When Probes See Errors that Confidence Missescited by 1

Built on by

Not cited yet.