MICCAI 2024 Papers — Page 2
International Conference on Medical Image Computing and Computer-Assisted Intervention · 856 papers
Binary Noise for Binary Tasks: Masked Bernoulli Diffusion for Unsupervised Anomaly Detection
Wolleb, Julia (University of Basel), Cattin, Philippe C. (University of Basel)
Anomaly DetectionDiffusion modelAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework for unsupervised medical anomaly detection based on a binary latent diffusion model is proposed, utilizing a binarized autoencoder to compress images into latent space, followed by Bernoulli diffusion recovery in that space, and improving anomaly mapping through a flipping probability for masking.
Biomechanics-informed Non-rigid Medical Image Registration and its Inverse Material Property Estimation with Linear and Nonlinear Elasticity
Min, Zhe (Shandong University), Hu, Yipeng (King's College London)
Point CloudBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Using Physics-Informed Neural Networks (PINN) to achieve non-rigid point set registration of soft tissues and material property estimation in both forward and inverse problems, and comparing linear and nonlinear elastic models.
Biophysics Informed Pathological Regularisation for Brain Tumour Segmentation
Zhang, Lipei (University of Cambridge), Aviles-Rivero, Angelica I (University of Cambridge)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: This paper proposes a pluggable biophysical information regularization method for brain tumor segmentation by combining a partial differential equation model of brain tumor growth with deep learning.
Biophysics-based data assimilation of longitudinal tau and amyloid-β PET scans
Wen, Zheyu (University of Texas at Austin), Biros, George (University of Texas at Austin)
Graph Neural NetworkTime SeriesBiomedical DataMagnetic Resonance ImagingPositron Emission TomographyAlzheimer's DiseaseOrdinary Differential Equation
🎯 What it does: A coupled Tau–Amyloidβ biophysical model is proposed, and model parameters and initial conditions are estimated through an inverse data assimilation method to fit multi-time-point PET scan data.
Black-Box Adaptation for Medical Image Segmentation
Paranjape, Jay N. (Johns Hopkins University), Patel, Vishal M. (Johns Hopkins University)
SegmentationTransformerPrompt EngineeringSimultaneous Localization and MappingImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The BAPS (Black-Box Adapter for Prompted Segmentation) method is proposed, which enhances the performance of medical image segmentation through visual prompts by utilizing a pre-trained image encoder and a trainable image-prompt decoder without accessing the underlying model parameters.
Blind Proximal Diffusion Model for Joint Image and Sensitivity Estimation in Parallel MRI
Li, Xing (Xian Jiaotong University), Xu, Zongben (Xian Jiaotong University)
RestorationDiffusion modelScore-based ModelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a blind proximal diffusion model (BPDM-PMRI) for the joint estimation and reconstruction of images and multi-channel sensitivity maps in parallel MRI.
Boosting FFPE-to-HE Virtual Staining with Cell Semantics from Pretrained Segmentation Model
Hu, Yihuang (Xiamen University), Wang, Liansheng (Shanghai Changhai Hospital)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a cGAN framework for enhancing FFPE-to-HE virtual staining using cell semantics extracted from a pre-trained cell segmentation model.
BPaCo: Balanced Parametric Contrastive Learning for Long-tailed Medical Image Classification
Cai, Zhiyuan (Southern University of Science and Technology), Tang, Xiaoying (Chinese University of Hong Kong)
ClassificationContrastive LearningImage
🎯 What it does: This paper proposes the BPaCo framework, which combines contrastive learning to achieve end-to-end training for the long-tail classification problem in medical images.
Brain Cortical Functional Gradients Predict Cortical Folding Patterns via Attention Mesh Convolution
Yang, Li, Zhang, Tuo (Northwestern Polytechnic University)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study predicts cortical folding patterns (i.e., brain folds) using a brain functional gradient through an attention grid convolution model.
Brain-Shift: Unsupervised Pseudo-Healthy Brain Synthesis for Novel Biomarker Extraction in Chronic Subdural Hematoma
Imre, Baris (University of Twente), Wolterink, Jelmer M. (University of Twente)
ClassificationSegmentationData SynthesisConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: A quantitative method for cSDH brain displacement based on unsupervised differential fractal pseudo-healthy brain synthesis is proposed, and this method is used to extract new biomarkers for predicting whether patients need surgery.
BrainSCK: Brain Structure and Cognition Alignment via Knowledge Injection and Reactivation for Diagnosing Brain Disorders
Wang, Lilong (Shanghai AI Laboratory), Wang, Xiaosong (Shanghai Artificial Intelligence Laboratory)
TransformerPrompt EngineeringVision Language ModelImageTextBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: The BrainSCK framework is proposed, which aligns brain structural images with cognitive knowledge using a two-stage visual-language model to achieve diagnosis of brain diseases across the lifespan;
BrainWaveNet: Wavelet-based Transformer for Autism Spectrum Disorder Diagnosis
Jeong, Ah-Yeong (Korea University), Suk, Heung-Il (Korea University)
ClassificationTransformerTime SeriesBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes BrainWaveNet, a waveform network based on Continuous Wavelet Transform (CWT) and Transformer for the diagnosis of Autism Spectrum Disorder (ASD).
Cache-Driven Spatial Test-Time Adaptation for Cross-Modality Medical Image Segmentation
Li, Xiang (Nanyang Technological University), Xu, Yanwu (South China University of Technology)
SegmentationDomain AdaptationBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a cache-driven Spatial Test-Time Adaptation (STTA) method to address the domain shift problem in cross-modal medical image segmentation.
Can Crowdsourced Annotations Improve AI-based Congestion Scoring For Bedside Lung Ultrasound?
Asgari-Targhi, Ameneh (Brigham and Women's Hospital and Harvard Medical School), Kapur, Tina (Brigham and Women's Hospital and Harvard Medical School)
SegmentationConvolutional Neural NetworkImageBiomedical DataUltrasound
🎯 What it does: Rapidly collected over 30k frames of B-line annotations through gamified crowdsourcing, trained an Attention U-Net to achieve automatic segmentation and counting of lung ultrasound B-lines, and explored the impact of data volume on accuracy.
Can LLMs’ Tuning Methods Work in Medical Multimodal Domain?
Chen, Jiawei (Fudan University), Zhang, Lihua (Fudan University)
TransformerLarge Language ModelVision Language ModelMultimodalityBiomedical Data
🎯 What it does: This paper transfers the parameter-efficient fine-tuning (PEFT) methods of LLM to medical multimodal vision-language models, constructing the MILE framework and systematically evaluating the effects of techniques such as LoRA, Prefix, IA3, and P-Tuning v2 on medical VLMs.
CAPTURE-GAN: Conditional Attribute Preservation through Unveiling Realistic GAN for artifact removal in dual-energy CT imaging
Park, Chunsu, Kim, MinWoo (Pusan National University)
RecognitionImage TranslationRestorationAnomaly DetectionConvolutional Neural NetworkGenerative Adversarial NetworkImageComputed Tomography
🎯 What it does: This paper proposes a cross-domain transformation framework based on Generative Adversarial Networks (GAN), mapping real images and forged images to the same feature space, and enhancing the robustness of facial anti-spoofing detection through adversarial, cycle consistency, identity preservation, and classification losses in joint training.
Car-Dcros: A Dataset and Benchmark for Enhancing Cardiovascular Artery Segmentation through Disconnected Components Repair and Open Curve Snake
Wang, Yuli (Johns Hopkins University), Bai, Harrison (Brown University)
SegmentationData SynthesisImageBiomedical DataComputed TomographyBenchmark
🎯 What it does: Developed an improved method for CTA vascular segmentation based on data-driven disconnection repair and open curve snake models.
CAR-MFL: Cross-Modal Augmentation by Retrieval for Multimodal Federated Learning with Missing Modalities
Poudel, Pranav (NepAl Applied Mathematics and Informatics Institute for research), Bhattarai, Binod (University of Aberdeen)
Federated LearningSafty and PrivacyConvolutional Neural NetworkTransformerImageTextMultimodalityElectronic Health RecordsRetrieval-Augmented Generation
🎯 What it does: A cross-modal retrieval-enhanced framework CAR-MFL is proposed, which utilizes a small amount of public multimodal data to retrieve and complete missing modalities in multimodal federated learning while maintaining data privacy.
Cardiac Copilot: Automatic Probe Guidance for Echocardiography with World Model
Jiang, Haojun (Tsinghua University), Huang, Gao (Tsinghua University)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkTransformerWorld ModelImageBiomedical DataUltrasound
🎯 What it does: Developed the Cardiac Copilot system, which utilizes the world model Cardiac Dreamer to predict six-dimensional movement commands for the probe on real-time ultrasound images, assisting non-professional medical personnel in completing cardiac ultrasound plane localization.
Cardiac Physiology Knowledge-driven Diffusion Model for Contrast-free Synthesis Myocardial Infarction Enhancement
Qi, Ronghui, Xu, Chenchu (Anhui University)
Image TranslationGenerationData SynthesisExplainability and InterpretabilityDiffusion modelGenerative Adversarial NetworkOptical FlowImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A cardiac physiology knowledge-driven diffusion model (CPKDM) is proposed for synthesizing myocardial infarction enhanced images (MIE) under non-contrast conditions, achieving high-quality synthesis by extracting kinematic and morphological features through a dual-stream structure.
CardioSpectrum: Comprehensive Myocardium Motion Analysis with 3D Deep Learning and Geometric Insights
Zuler, Shahar, Raviv, Dan (Tel Aviv University)
SegmentationOptimizationConvolutional Neural NetworkOptical FlowImageBiomedical DataComputed Tomography
🎯 What it does: This study proposes a complete framework that combines 3D deep learning and Functional Maps to accurately capture the three-dimensional motion of the left ventricular myocardium from cardiac CT angiography, with a particular focus on small tangential displacements.
Cardiovascular Disease Detection from Multi-View Chest X-rays with BI-Mamba
Yang, Zefan (Rensselaer Polytechnic Institute), Yan, Pingkun (Rensselaer Polytechnic Institute)
ClassificationAnomaly DetectionComputational EfficiencyConvolutional Neural NetworkTransformerImageMultimodalityComputed Tomography
🎯 What it does: This study investigates the use of multi-view chest X-rays to predict cardiovascular disease risk.
Causal Intervention for Brain tumor Segmentation
Liu, Hengxin (Tianjin University), Liu, Anan (Tianjin University)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A brain tumor segmentation model based on causal intervention is proposed, which explicitly eliminates the interference of background and different tumor categories using front-door adjustment and regional/category causal modules, achieving more accurate segmentation.
CausalCLIPSeg: Unlocking CLIP’s Potential in Referring Medical Image Segmentation with Causal Intervention
Chen, Yaxiong (Wuhan University of Technology), Mou, Lichao (Xidian University)
SegmentationConvolutional Neural NetworkContrastive LearningImageTextMultimodalityComputed Tomography
🎯 What it does: CausalCLIPSeg is proposed, an end-to-end multimodal medical image segmentation framework that utilizes CLIP's visual and text encoders and achieves pixel-level text-image alignment through cross-modal decoding, while introducing a causal intervention module to eliminate confounding bias.
Causality-Informed Fusion Network for Automated Assessment of Parkinsonian Body Bradykinesia
Quan, Yuyang (Shanghai Jiao Tong University), Qian, Xiaohua (Shanghai Jiao Tong University)
ClassificationRecognitionGraph Neural NetworkVideoMultimodality
🎯 What it does: This paper proposes a video-based automatic assessment framework for body bradykinesia in Parkinson's disease, using only consumer-grade cameras. It utilizes videos of gait and leg agility movements, employing Graph Convolutional Networks (GCN) to extract skeletal motion features, and achieves a three-class assessment through causal information-guided feature fusion.
CausCLIP: Causality-Adapting Visual Scoring of Visual Language Models for Few-Shot Learning in Portable Echocardiography Quality Assessment
Li, Yiran (Shandong University), Li, Shuo (Harbin Institute of Technology)
TransformerVision Language ModelContrastive LearningImageBiomedical DataUltrasound
🎯 What it does: This paper proposes CausCLIP, which utilizes a visual-language model combined with causal learning for transfer learning in few-shot portable echocardiogram quality assessment.
CAVM: Conditional Autoregressive Vision Model for Contrast-Enhanced Brain Tumor MRI Synthesis
Gui, Lujun (Beijing Institute of Technology), Yan, Tianyi (Beijing Institute of Technology)
SegmentationGenerationData SynthesisTransformerGenerative Adversarial NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework based on a Conditional Autoregressive Visual Model (CAVM) is proposed to generate contrast-enhanced T1Gd images from non-contrast brain tumor MRI (T1w, T2w, FLAIR), achieving high-quality synthesis through a stepwise increasing dosage.
ccRCC Metastasis Prediction via Exploring High-Order Correlations on Multiple WSIs
Zhou, Huijian (Xi'an Jiaotong University), Gao, Yue (Tsinghua University)
ClassificationGraph Neural NetworkImageBiomedical Data
🎯 What it does: A transfer prediction model based on Multi-Slice Hypergraph Convolution (MSHGC) is proposed to utilize high-order intra- and inter-related features in multiple whole slide images (WSI) of patients with clear cell renal cell carcinoma (ccRCC) for metastatic risk scoring.
Center-to-Edge Denoising Diffusion Probabilistic Models with Cross-domain Attention for Undersampled MRI Reconstruction
Zhao, Jianfeng (Western University), Li, Shuo (Harbin Institute of Technology)
RestorationDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A center-to-edge (C2E) denoising diffusion probabilistic model (C2E-DDPM) is proposed, which combines cross-domain attention to achieve undersampling reconstruction of fully sampled MRI.
Centerline Boundary Dice Loss for Vascular Segmentation
Shi, Pengcheng (Harbin Institute of Technology (Shenzhen)), Ma, Ting (Harbin Institute of Technology at Shenzhen)
SegmentationConvolutional Neural NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A centerline boundary-based Dice loss function (cbDice) is proposed to improve the topological preservation and geometric detail capture in vascular segmentation.
Centerline-Diameters Data Structure for Interactive Segmentation of Tube-shaped Objects
Sirazitdinov, Ilyas (Skolkovo Institute of Science and Technology), Dylov, Dmitry V. (Skolkovo Institute of Science and Technology)
SegmentationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A new data structure and click encoding based on centerlines and diameters are proposed, enabling interactive tubular object segmentation without relying on binary masks.
Cephalometric Landmark Detection across Ages with Prototypical Network
Wu, Han (ShanghaiTech University), Cui, Zhiming (ShanghaiTech University)
RecognitionObject DetectionConvolutional Neural NetworkImage
🎯 What it does: The CeLDA method is proposed, which uses prototypical networks to achieve landmark detection in lateral skull X-ray images across ages (adolescents and adults), and improves detection accuracy through global prototypes, prototype alignment, and mask prototype relationship mining.
Characterizing the left ventricular ultrasound dynamics in the frequency domain to estimate the cardiac function
Carrera-Pinzón, Andrés Felipe (Universidad Nacional De Colombia), Iregui Guerrero, Marcela (Universidad Nacional De Colombia)
VideoBiomedical DataUltrasound
🎯 What it does: By performing frequency domain analysis of the left ventricular area over time in four-chamber view ultrasound videos, new cardiac function indicators are proposed.
CheXtriev: Anatomy-Centered Representation for Case-Based Retrieval of Chest Radiographs
Akash R. J., Naren (International Institute of Information Technology Hyderabad), Sivaswamy, Jayanthi (International Institute of Information Technology Hyderabad)
RetrievalGraph Neural NetworkTransformerImageBiomedical Data
🎯 What it does: The CheXtriev framework is proposed, utilizing a graph Transformer to extract features from 18 anatomical regions and achieve case retrieval of chest X-ray images.
CINA: Conditional Implicit Neural Atlas for Spatio-Temporal Representation of Fetal Brains
Dannecker, Maik (Technical University Munich), Rueckert, Daniel (Technical University of Munich)
SegmentationGenerationAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A method for generating fetal brain atlases based on Conditional Implicit Neural Atlas (CINA) is proposed, which can generate continuous brain structure and tissue probability maps at any spatiotemporal resolution.
Class and Region-Adaptive Constraints for Network Calibration
Murugesan, Balamurali (ETS Montreal), Dolz, Jose (CHU Sainte-Justine, University of Montreal)
SegmentationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImageBiomedical DataMagnetic Resonance ImagingComputed TomographyBenchmark
🎯 What it does: This paper proposes a Class and Region Adaptive Constraint (CRaC), which introduces class and region-specific penalty terms into the semantic segmentation network and adaptively learns these penalty weights during training using the Augmented Lagrangian Method, thereby achieving network calibration.
Class-aware Mutual Mixup with Triple Alignments for Semi-Supervised Cross-domain Segmentation
Cai, Zhuotong (National Key Laboratory Of Human Machine Hybrid Augmented Intelligence), Duncan, James S. (Yale University)
SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A Class-aware Mutual Mixup and Triple Alignments method for semi-supervised cross-domain segmentation is proposed, achieving synchronous alignment and data diversification between the source domain, labeled target domain, and unlabeled target domain.
Class-Balancing Deep Active Learning with Auto-Feature Mixing and Minority Push-Pull Sampling
Lin, Hongxin (Shenzhen University), Huang, Bingsheng (Shenzhen University)
ClassificationComputational EfficiencyConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A class-balanced deep active learning framework for medical image classification, CB-DAL, is proposed, which combines two modules, Auto-FM and MPPS, for sample selection.
CLEFT: Language-Image Contrastive Learning with Efficient Large Language Model and Prompt Fine-Tuning
Du, Yuexi (Yale University), Dvornek, Nicha C. (Yale University)
ClassificationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes the CLEFT framework, which combines pre-trained LLMs with parameter-efficient fine-tuning (PEFT) for contrastive learning in medical images, and enhances model generalization through context prompt learning after pre-training.
Clinical-grade Multi-Organ Pathology Report Generation for Multi-scale Whole Slide Images via a Semantically Guided Medical Text Foundation Model
Tan, Jing Wei (Korea University), Jeong, Won-Ki (Korea University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: A pathology report generation model PMPRG based on a multi-scale regional visual Transformer (MR-ViT) is proposed, which can automatically generate clinical-grade pathology reports for multiple organs and multiple slices.
CLIP-DR: Textual Knowledge-Guided Diabetic Retinopathy Grading with Ranking-aware Prompting
Yu, Qinkai (University of Exeter), Meng, Yanda (University Of Exeter)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageBenchmark
🎯 What it does: A CLIP-based framework for diabetic retinopathy (DR) grading, named CLIP-DR, is proposed, which utilizes text prompts to learn the natural order of DR grading and addresses the long-tail data problem.
Coarse-Grained Mask Regularization for Microvascular Obstruction Identification from non-contrast Cardiac Magnetic Resonance
Yan, Yige (Nanyang Technological University), Rajapakse, Jagath C. (Nanyang Technological University)
RecognitionConvolutional Neural NetworkTransformerImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework for identifying microvascular obstruction (MVO) based on contrast-free cardiac CINE MRI is proposed, enhancing model learning through coarse-grained mask regularization.
Coarse-to-Fine Latent Diffusion Model for Glaucoma Forecast on Sequential Fundus Images
Zhang, Yuhan (Chinese University of Hong Kong), Heng, Pheng-Ann (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelImageSequentialBiomedical Data
🎯 What it does: Based on continuous retinal fluorescence imaging, a coarse-to-fine latent diffusion model (C2F-LDM) is used to generate latent features for future time points, thereby predicting glaucoma risk and reconstructing future fundus images.
CoBooM: Codebook Guided Bootstrapping for Medical Image Representation Learning
Singh, Azad (Indian Institute of Technology), Mishra, Deepak (Government Medical College Srinagar)
ClassificationSegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a codebook-based self-supervised learning framework called CoBooM to enhance representation learning in medical images.
Common Vision-Language Attention for Text-Guided Medical Image Segmentation of Pneumonia
Guo, Yunpeng (Sichuan University), Wang, Yan (East China Normal University)
SegmentationConvolutional Neural NetworkVision Language ModelImageTextMultimodalityMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A text-guided common attention model (TGCAM) is proposed for the automatic segmentation of infection areas in pneumonia images.
Comprehensive Generative Replay for Task-Incremental Segmentation with Concurrent Appearance and Semantic Forgetting
Li, Wei (Shanghai Jiao Tong University), Gu, Lixu (Shanghai Jiao Tong University)
SegmentationGenerationDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A Task-Incremental Learning (TIL) framework called Comprehensive Generative Replay (CGR) is proposed, which generates image-mask pairs of past tasks at each step to recover lost appearance and semantic knowledge, while simultaneously updating the model to support future replay during new task learning.
Concept-Attention Whitening for Interpretable Skin Lesion Diagnosis
Hou, Junlin (Hong Kong University of Science and Technology), Chen, Hao (Harvard Medical School)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: An interpretable skin disease diagnosis model named Concept-Attention Whitening (CAW) has been developed, which aligns the axes of the latent space with medical concepts by inserting whitening layers and orthogonal transformations in the feature encoder, thereby enabling the interpretation of diagnostic features.
Conditional 4D Motion Diffusion Models with Masked Observations to Forecast Deformations
Thibeault, Sylvain (Polytechnique Montreal), Kadoury, Samuel (CHUM Hospital Research Center)
Diffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a 4D deformation prediction method based on a conditional diffusion model, utilizing 2D real-time MR images to predict the future deformation field of the liver.
Conditional Diffusion Model for Versatile Temporal Inpainting in 4D Cerebral CT Perfusion Imaging
Bae, Juyoung (Hong Kong University of Science and Technology), Chen, Hao (Harvard Medical School)
RestorationData SynthesisConvolutional Neural NetworkDiffusion modelBiomedical DataComputed Tomography
🎯 What it does: Using conditional diffusion models for temporal interpolation of brain CT perfusion sequences to fill in missing scans at high temporal resolution, outputting synthetic scans at one-second intervals.
Conditional diffusion model with spatial attention and latent embedding for medical image segmentation
Hejrati, Behzad (Wayne State University), Dong, Ming (Wayne State University)
SegmentationConvolutional Neural NetworkDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A conditional diffusion model cDAL is proposed, utilizing spatial attention and random latent embeddings for medical image segmentation.
Conditional Score-Based Diffusion Model for Cortical Thickness Trajectory Prediction
Xiao, Qing (Southern Medical University), Li, Xiang (Nanyang Technological University)
Recurrent Neural NetworkDiffusion modelScore-based ModelTime SeriesBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: A framework for predicting the temporal changes in cortical thickness based on a conditional score diffusion model is proposed, which can continuously predict the cortical thickness trajectory of Alzheimer's disease patients solely based on baseline information (age, gender, diagnosis, previous cortical thickness, and the time interval between two scans) and provide uncertainty estimates for the predictions.
Confidence intervals uncovered: Are we ready for real-world medical imaging AI?
Christodoulou, Evangelia (German Cancer Research Center), Maier-Hein, Lena (Dresden University of Technology)
SegmentationBiomedical DataReview/Survey PaperBenchmark
🎯 What it does: The system evaluates and summarizes the reporting of performance variability in medical image segmentation models in MICCAI 2023 papers, and estimates the unreported standard deviations and confidence intervals using polynomial models.
Confidence Matters: Enhancing Medical Image Classification Through Uncertainty-Driven Contrastive Self-Distillation
Sharma, Saurabh (Indian Institute of Technology Patna), Chandra, Joydeep (Indian Institute of Technology Patna)
ClassificationKnowledge DistillationContrastive LearningImageBiomedical Data
🎯 What it does: A framework for uncertainty-driven contrastive self-distillation (UDCD) is proposed to address issues of data scarcity, class imbalance, and high intra-class variance in medical image classification.
Confidence-guided Semi-supervised Learning for Generalized Lesion Localization in X-ray Images
Das, Abhijit (Jio Institute), Roy, Sudipta (Inception Institute of AI)
Object DetectionSegmentationKnowledge DistillationImageBiomedical Data
🎯 What it does: A confidence-guided semi-supervised framework called AnoMed is proposed for multi-scale lesion localization in chest X-rays, addressing the issues of pseudo-label inconsistency and small class bias by combining Scale-Invariant Bottleneck (SIB) and Confidence-Guided Pseudo-Label Optimizer (PLO).
Consecutive-Contrastive Spherical U-net: Enhancing Reliability of Individualized Functional Brain Parcellation for Short-duration fMRI Scans
Hu, Dan (University of North Carolina at Chapel Hill), Li, Gang (University of North Carolina at Chapel Hill)
SegmentationConvolutional Neural NetworkContrastive LearningBiomedical DataMagnetic Resonance Imaging
🎯 What it does: By constructing a continuous contrastive learning spherical U-Net model, this study predicts individualized brain region segmentation from short-term fMRI scans, thereby enhancing the reliability and usability of short-term scans.
Context-guided Continual Reinforcement Learning for Landmark Detection with Incomplete Data
Wan, Kaiwen (Fudan University), Zhuang, Xiahai (Fudan University)
Object DetectionReinforcement LearningPrompt EngineeringImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper studies a context-guided continuous reinforcement learning framework (CgCRL) for multi-point label detection in sequentially incomplete medical imaging data.
Continual Domain Incremental Learning for Privacy-aware Digital Pathology
Kumari, Pratibha (University of Regensburg), Merhof, Dorit (Hannover Medical School)
Object DetectionSafty and PrivacyConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningBiomedical Data
🎯 What it does: This study investigates a privacy-friendly incremental learning method in continuous domains for tumor detection in digital pathology, avoiding the storage of historical samples.
Continually Tuning a Large Language Model for Multi-domain Radiology Report Generation
Sun, Yihua (Tsinghua University), Liao, Hongen (Tsinghua University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningTextMultimodalityBiomedical DataComputed TomographyElectronic Health Records
🎯 What it does: A continuous learning framework based on LLM is proposed, utilizing minimal learnable parameters to achieve no-forgetting performance in multi-center, multi-modal radiology report generation tasks.
Contrast Representation Learning from Imaging Parameters for Magnetic Resonance Image Synthesis
Xiong, Honglin (ShanghaiTech University), Wang, Qian (United Imaging Intelligence)
GenerationData SynthesisRepresentation LearningAuto EncoderContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A contrast representation learning (CRL) method based on imaging parameters is proposed for synthesizing images with different contrasts in magnetic resonance imaging.
Controllable and Efficient Multi-Class Pathology Nuclei Data Augmentation using Text-Conditioned Diffusion Models
Oh, Hyun-Jic (Korea University), Jeong, Won-Ki (Korea University)
ClassificationSegmentationData SynthesisDiffusion modelImageBiomedical Data
🎯 What it does: A two-stage text-conditioned diffusion model framework is designed for the label and image generation of multi-class pathological nuclei data, achieving controllable high-quality data augmentation.
Controllable Counterfactual Generation for Interpretable Medical Image Classification
Liu, Shiyu (Xi'an Jiaotong University), Ma, Jianhua (Pazhou Lab)
ClassificationGenerationData SynthesisExplainability and InterpretabilityDiffusion modelImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: This paper proposes a controllable counterfactual generation method based on diffusion models to generate image differences between different pathological categories (such as Alzheimer's disease and healthy normals), thereby providing interpretable diagnostic results and performing data augmentation.
Convex Segments for Convex Objects using DNN Boundary Tracing and Graduated Optimization
Pal, Jimut B. (Indian Institute of Technology Bombay), Awate, Suyash P. (Indian Institute of Technology Bombay)
SegmentationOptimizationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A deep neural network framework is proposed that ensures the segmentation results of convex objects in two-dimensional images are convex, utilizing geometric boundary constraints to achieve hard convexity constraints.
Convolutional Implicit Neural Representation of pathology whole-slide images
Lee, DongEon (Pusan National University), Kim, MinWoo (Pusan National University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A Convolutional Implicit Neural Representation (CINR) model is proposed for the reconstruction of high-resolution whole-slide pathology images, enhancing high-frequency detail recovery by integrating multi-resolution hash encoding and CNN.
CoReEcho: Continuous Representation Learning for 2D+time Echocardiography Analysis
Maani, Fadillah Adamsyah (Mohamed bin Zayed University of Artificial Intelligence), Yaqub, Mohammad (Mohamed bin Zayed University of Artificial Intelligence)
Representation LearningTransformerContrastive LearningVideoBiomedical DataUltrasound
🎯 What it does: Proposes the CoReEcho framework, achieving direct ejection fraction regression from 2D+time cardiac ultrasound videos through continuous representation learning.
Correlation-adaptive Multi-view CEUS Fusion for Liver Cancer Diagnosis
Wan, Peng (Nanjing University of Aeronautics and Astronautics), Zhang, Daoqiang (Nanjing University of Aeronautics and Astronautics)
ClassificationBiomedical DataUltrasound
🎯 What it does: A correlation adaptive method CAMVF based on multi-view CEUS fusion is proposed for the diagnosis of hepatocellular carcinoma and intrahepatic cholangiocarcinoma.
Cortical Surface Reconstruction from 2D MRI with Segmentation-Constrained Super-Resolution and Representation Learning
Wu, Wenxuan (South China University of Technology), Zhang, Xin (South China University of Technology)
SegmentationSuper ResolutionRepresentation LearningConvolutional Neural NetworkTransformerImageBiomedical DataMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: A two-stage process for reconstructing the complete cortical surface from 2D low-resolution MRI is proposed, including super-resolution with segmentation constraints and surface reconstruction with feature alignment.
COVID19 to Pneumonia: Multi Region Lung Severity Classification using CNN Transformer Position-Aware Feature Encoding Network
Lee, Jong Bub (Inha University), Lee, Hyun Gyu (Inha University)
ClassificationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a multi-region lung severity classification method based on a CNN-Transformer hybrid network, utilizing position-aware feature encoding and region-shared MLP to achieve automatic scoring of COVID-19 and heterogeneous pneumonia images.
CP-CLIP: Core-Periphery Feature Alignment CLIP for Zero-Shot Medical Image Analysis
Yu, Xiaowei (University of Texas at Arlington), Zhu, Dajiang (University of Georgia)
ClassificationSegmentationExplainability and InterpretabilityGraph Neural NetworkContrastive LearningImageTextMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: An improved CLIP model is proposed for medical imaging and corresponding clinical reports, utilizing a Core-Periphery (CP) graph-guided neural network to achieve efficient alignment of image and text features, thereby enabling zero-shot medical image classification and lesion region localization.
CriDiff: Criss-cross Injection Diffusion Framework via Generative Pre-train for Prostate Segmentation
Liu, Tingwei (Dalian University of Technology), Lu, Huchuan (Dalian University of Technology)
SegmentationTransformerDiffusion modelImageBiomedical DataMagnetic Resonance ImagingUltrasound
🎯 What it does: A two-stage Criss-cross Injection Diffusion (CriDiff) framework is proposed, combining Boundary Enhancement (BEC) and Core Enhancement (CEC) conditioners with Generative Pre-training (GP) methods for prostate segmentation.
Cross Prompting Consistency with Segment Anything Model for Semi-supervised Medical Image Segmentation
Miao, Juzheng (Chinese University of Hong Kong), Heng, Pheng-Ann (Chinese University of Hong Kong)
SegmentationSupervised Fine-TuningImageBiomedical DataMagnetic Resonance ImagingUltrasound
🎯 What it does: A cross-prompt consistency semi-supervised medical image segmentation method based on the Segment Anything model, CPC-SAM, has been developed.
Cross-conditioned Diffusion Model for Medical Image to Image Translation
Xing, Zhaohu (Hong Kong University of Science and Technology (Guangzhou)), Zhu, Lei (Hong Kong University of Science and Technology)
Image TranslationData SynthesisDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a Cross-Conditional Diffusion Model (CDM) for modality conversion in medical imaging, capable of synthesizing missing T1c and T2f images in the absence of certain MRI modalities.
Cross-Dimensional Medical Self-Supervised Representation Learning Based on a Pseudo-3D Transformation
Gao, Fei (Peking University), Yu, Yizhou (University of Hong Kong)
ClassificationSegmentationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A cross-dimensional self-supervised learning framework based on pseudo-3D transformation is proposed, jointly pre-training 2D and 3D medical images.
Cross-graph Interaction and Diffusion Probability Models for Lung Nodule Segmentation
Su, Huaqiang (Shenzhen University), Lei, Baiying (Shenzhen University)
SegmentationConvolutional Neural NetworkTransformerDiffusion modelImageComputed Tomography
🎯 What it does: This paper proposes Diff-UNet, a denoising UNet that combines a dual-branch DPM, Transformer, and CNN to achieve high-precision segmentation of lung nodules in CT images.
Cross-modal Diffusion Modelling for Super-resolved Spatial Transcriptomics
Wang, Xiaofei (University of Cambridge), Li, Chao (University of Dundee)
RestorationSuper ResolutionDiffusion modelImageBiomedical Data
🎯 What it does: This paper proposes a cross-modal conditional diffusion model (Diff-ST) to achieve spatial transcriptomics (ST) super-resolution, utilizing tissue slice images and low-resolution ST data to jointly recover high-resolution ST maps.
Cross-Modality Cardiac Insight Transfer: A Contrastive Learning Approach to Enrich ECG with CMR Features
Ding, Zhengyao (Zhejiang University), Huang, Zhengxing (First Affiliated Hospital of Zhejiang University School of Medicine)
ClassificationRepresentation LearningTransformerAuto EncoderContrastive LearningMultimodalityTime SeriesBiomedical DataMagnetic Resonance ImagingElectrocardiogram
🎯 What it does: By using contrastive learning to transfer cardiac magnetic resonance (CMR) features to electrocardiogram (ECG) embeddings, ECG can more accurately predict cardiac structural indicators and disease states.
Cross-Phase Mutual Learning Framework for Pulmonary Embolism Identification on Non-Contrast CT Scans
Bai, Bizhe (DAMO Academy, Alibaba Group), Xu, Minfeng (Zhejiang University)
ClassificationSegmentationExplainability and InterpretabilityKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImageBiomedical DataComputed Tomography
🎯 What it does: This study proposes a Cross-Phase Mutual Learning framework (CPMN) that utilizes dual-phase scanning with CTPA and NCT for multi-task recognition of pulmonary embolism (PE), including classification and segmentation, and enhances diagnostic accuracy on NCT scans through knowledge transfer.
Cross-Slice Attention and Evidential Critical Loss for Uncertainty-Aware Prostate Cancer Detection
Hung, Alex Ling Yu (UCLA), Sung, Kyunghyun (UCLA)
Object DetectionSegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A cross-slice attention mechanism that utilizes global and local information (GLCSA) is proposed, combined with evidence key loss (EC Loss) for evidence deep learning, to achieve detection and uncertainty estimation of prostate cancer.
CryoSAM: Training-free CryoET Tomogram Segmentation with Foundation Models
Zhao, Yizhou (Carnegie Mellon University), Xu, Min (University of Alabama at Birmingham)
SegmentationPrompt EngineeringContrastive LearningImage
🎯 What it does: A training-free CryoSAM framework is proposed to complete the segmentation of all particles in CryoET using single-point prompts.
Cryotrack: Planning and Navigation for Computer Assisted Cryoablation
Krumb, Henry J. (Technische Universität Darmstadt), Essert, Caroline (University of Strasbourg)
OptimizationBiomedical DataComputed Tomography
🎯 What it does: Cryotrack is an integrated computer-assisted cryoablation system that provides preoperative automatic risk structure identification, feasible entry point map generation, and real-time needle path navigation with audio feedback during the procedure, helping doctors accurately insert the needle.
CS3: Cascade SAM for Sperm Segmentation
Shi, Yi (Nanjing University), Zhan, De-Chuan (Nanjing University)
SegmentationImage
🎯 What it does: This paper proposes an unsupervised CS3 algorithm that utilizes Cascade SAM to accurately segment overlapping sperm images through a three-stage progression: head separation, simple tail, and complex tail.
CT-based brain ventricle segmentation via diffusion Schrödinger Bridge without target domain ground truths
Teimouri, Reihaneh (Concordia University), Xiao, Yiming (Concordia University)
SegmentationDiffusion modelImageBiomedical DataComputed Tomography
🎯 What it does: This study proposes a CT brain ventricle segmentation method based on the diffusion Schrödinger bridge, which does not require target domain labeling.
CT2Rep: Automated Radiology Report Generation for 3D Medical Imaging
Hamamci, Ibrahim Ethem (University Of Zurich), Menze, Bjoern (University of Zurich)
GenerationTransformerImageTextMultimodalityComputed Tomography
🎯 What it does: This paper presents CT2Rep, which achieves the automatic generation of radiology reports from 3D chest CT images for the first time.
Curriculum Prompting Foundation Models for Medical Image Segmentation
Zheng, Xiuqi (Beijing University of Posts and Telecommunications), Lao, Qicheng (Ningbo Fregty Optoelectronics Technology Co., Ltd)
SegmentationPrompt EngineeringImageBiomedical DataUltrasound
🎯 What it does: This paper proposes a medical image segmentation framework based on curriculum-based prompts, utilizing self-generated multi-granularity prompts (boxes, points, masks) to achieve automated segmentation on the SAM model.
Customized Relationship Graph Neural Network for Brain Disorder Identification
Xia, Zhengwang (Nanjing University of Science and Technology), Lu, Jianfeng (Nanjing University of Science and Technology)
ClassificationRecognitionGraph Neural NetworkBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: A CRGNN model is proposed that can jointly learn brain network structure and classification tasks for brain disease recognition.
Cut to the Mix: Simple Data Augmentation Outperforms Elaborate Ones in Limited Organ Segmentation Datasets
Liu, Chang (Friedrich-Alexander-Universitt Erlangen-Nrnberg), Maier, Andreas (Friedrich-Alexander-Universität Erlangen)
SegmentationConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: A systematic evaluation and comparison of four algorithms based on image fusion and object-level augmentation (CutMix, CarveMix, ObjectAug, AnatoMix) was conducted on a limited multi-organ segmentation dataset, demonstrating that the simple CutMix is the most effective in improving segmentation accuracy.
CUTS: A Deep Learning and Topological Framework for Multigranular Unsupervised Medical Image Segmentation
Liu, Chen (Yale University), Krishnaswamy, Smita (Yale University)
SegmentationConvolutional Neural NetworkContrastive LearningImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: An unsupervised multi-granularity medical image segmentation framework called CUTS is proposed, which generates embeddings using contrastive learning and reconstruction loss based on pixel-centered patches, and achieves multi-scale segmentation through diffusion condensation.
Cycle-consistent Learning for Fetal Cortical Surface Reconstruction
Dong, Xiuyu (University of North Carolina), Li, Gang (University of North Carolina at Chapel Hill)
RestorationSegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: This paper proposes a multi-task deep learning framework based on cyclic consistency and differential homeomorphic deformation for the reconstruction of the inner and outer surfaces of fetal brain MRI.
D-CoRP: Differentiable Connectivity Refinement for Functional Brain Networks
Hu, Haoyu, Li, Chao (University of Dundee)
Graph Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: D-CoRP is proposed, a differentiable functional brain network connectivity refinement plugin that can automatically remove noise or redundant edges while maintaining the overall structure.
D-MASTER: Mask Annealed Transformer for Unsupervised Domain Adaptation in Breast Cancer Detection from Mammograms
Ashraf, Tajamul (Indian Institute of Technology Delhi), Arora, Chetan (PGIMER Chandigarh)
Object DetectionDomain AdaptationTransformerAuto EncoderImage
🎯 What it does: This paper proposes an unsupervised domain adaptation framework based on Transformer, named D-MASTER, specifically designed to address the cross-domain transfer problem of tumor detection in mammograms (breast X-rays). The framework combines a teacher-student structure with a Masked Autoencoder (MAE) for adaptive masking and reconstruction of multi-scale features, and utilizes adaptive confidence refinement to filter out pseudo-labels, ultimately achieving efficient detection of target domain breast images.
Data Augmentation with Multi-armed Bandit on Image Deformations Improves Fluorescence Glioma Boundary Recognition
Xiao, Anqi (Institute of Automation, Chinese Academy of Sciences), Hu, Zhenhua (Institute of Automation, Chinese Academy of Sciences)
RecognitionSegmentationData-Centric LearningConvolutional Neural NetworkReinforcement LearningImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper proposes an adaptive data augmentation method based on multi-armed bandits (MAB) called EEA, aimed at enhancing the generalization ability of deep learning models for brain glioma boundary recognition under low sample size NIR-II fluorescence imaging, and it is validated on three modalities of data: white light, NIR-I, and NIR-II.
Data-Algorithm-Architecture Co-Optimization for Fair Neural Networks on Skin Lesion Dataset
Sheng, Yi (George Mason University), Yang, Lei (University of North Carolina at Chapel Hill)
OptimizationData-Centric LearningNeural Architecture SearchConvolutional Neural NetworkReinforcement LearningImageBiomedical Data
🎯 What it does: Proposed and implemented the BiaslessNAS framework, which uses a reinforcement learning controller to simultaneously optimize data sampling, training algorithms, and network architectures on skin lesion datasets to enhance the fairness and accuracy of the model.
Data-Driven Tissue- and Subject-Specific Elastic Regularization for Medical Image Registration
Reithmeir, Anna (Technical University of Munich), Zimmer, Veronika A. (Technical University of Munich)
Convolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A medical image registration method based on hypernetworks is proposed, which can learn patient- and tissue-specific linear elastic regularization parameters without retraining.
DB-SAM: Delving into High Quality Universal Medical Image Segmentation
Qin, Chao (Mohamed bin Zayed University of Artificial Intelligence), Anwer, Rao Muhammad (Mohamed Bin Zayed University of Artificial Intelligence)
SegmentationTransformerMultimodalityBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasound
🎯 What it does: Designed and implemented the DB-SAM dual-branch (ViT + convolution) framework, transforming SAM to achieve unified high-quality medical image segmentation.
DCDiff: Dual-Domain Conditional Diffusion for CT Metal Artifact Reduction
Shen, Ruochong (Monash University), Ke, Qiuhong (Monash University)
RestorationDiffusion modelImageComputed Tomography
🎯 What it does: This paper proposes a dual-domain conditional diffusion framework (DCDiff) based on diffusion models for the removal of metal artifacts in CT images.
DCrownFormer: Morphology-aware Point-to-Mesh Generation Transformer for Dental Crown Prosthesis from 3D Scan Data of Antagonist and Preparation Teeth
Yang, Su (Seoul National University), Yi, Won-Jin (Seoul National University)
RestorationGenerationTransformerPoint CloudMesh
🎯 What it does: Directly generate crown meshes from 3D scanned point clouds of opposing teeth and prepared teeth using the Transformer architecture.
Death by Retrospective Undersampling - Caveats and Solutions for Learning-Based MRI Reconstructions
Rajput, Junaid R. (University Hospital Erlangen), Zaiss, Moritz (Friedrich-Alexander-University Erlangen-Nürnberg)
RestorationBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper studies the impact of retrospective undersampling on signal dynamics in MRI learning-based reconstruction, demonstrating its errors through differentiable Bloch simulations, while providing the MR-zero tool to generate correct training data.
Debiased Noise Editing on Foundation Models for Fair Medical Image Classification
Jin, Ruinan (University of British Columbia), Li, Xiaoxiao (University Of British Columbia)
ClassificationTransformerSupervised Fine-TuningImageBiomedical Data
🎯 What it does: This paper proposes a method to eliminate false associations related to sensitive attributes by adding learnable denoising noise (DNE) to medical images, and implements fair embedding in a black-box base model API.
Decoding the visual attention of pathologists to reveal their level of expertise
Chakraborty, Souradeep (Stony Brook University), Samaras, Dimitris (Stony Brook University)
TransformerImageBiomedical Data
🎯 What it does: The study utilizes the gaze trajectories of pathologists when reading whole slide images of prostate tissue to establish a model that predicts their visual attention heat maps, thereby assessing their professional level.
Decoupled Training for Semi-supervised Medical Image Segmentation with Worst-Case-Aware Learning
Das, Ankit (Agency for Science, Technology and Research), Liu, Yong (Beijing University of Posts and Telecommunications)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The paper proposes a method to improve semi-supervised medical image segmentation through decoupled training and worst-case aware learning.
Deep intra-operative illumination calibration of hyperspectral cameras
Baumann, Alexander (Siemens AG), Maier-Hein, Lena (Dresden University of Technology)
SegmentationData SynthesisConvolutional Neural NetworkAuto EncoderImageBiomedical Data
🎯 What it does: A real-time illumination calibration method based on deep learning has been proposed and implemented in the operating room environment, capable of correcting the spatial resolution of hyperspectral images without the need for manual whiteboard calibration.