Mingjie Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mingjie Li has a total of 77 co-authors. Here are the top co-authors by number of shared publications:
- Xiaojun Chang
University of Science and Technology of China, Hefei, China
9 shared publications - Junfei Yi
Hunan University
4 shared publications Xiaojun Chang
University of Science and Technology of China
4 shared publications- Hui Zhang
Hunan University
3 shared publications - Jianxu Mao
Hunan University
3 shared publications - Junbin Gao
University of Sydney
3 shared publications
Latest Publications
- Open Access
- 8 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
Curriculum Learning With Chaos-Aware Gating Mechanism for Infant Cardiorespiratory Status Prediction via Heart Rate Variability and Bedside Data
Predicting the cardiorespiratory status of extremely preterm infants using signals from bedside monitors is critical for timely intervention in neonatal intensive care units (NICUs), potentially improving long-term outcomes. However, directly adapting existing time series data analysis methods to this task is challenged by the scarcity of neonatal data and the high noise levels inherent to physiological signals in this population. To address these challenges, we propose a novel hybrid neural network that integrates C urriculum L earning and a C haos-aware G ating mechanism, called CLCG. Given the long multivariate physiological sequences, we first apply a sliding-window strategy to segment the data into overlapping windows. For each window, a shared Convolutional Neural Network (CNN) fuses multi-modal signals and extracts localized representations. To increase data diversity and enhance generalization capability, we introduce a curriculum learning-based data augmentation scheme that progressively reduces the number of windows during training. This encourages the model to learn discriminative representations by exposing it to increasingly challenging scenarios with less temporal context. To further mitigate the impact of signal noise and improve temporal modeling, we extract handcrafted features, the Lyapunov exponent, from the heart rate variability (HRV) parameters. These features are incorporated into a novel chaos-aware gate mechanism within the Long Short-Term Memory (LSTM) network. This mechanism guides the gating process based on nonlinear dynamical properties, and suppresses noise-related fluctuations. Finally, a Multi-Layer Perceptron (MLP) predicts the respiratory and hypotensive status over the next five hours. We evaluate our model on two datasets: an in-house cohort of 55 extremely preterm infants from the Monash Children's Hospital, and the publicly available MC-MED dataset from emergency department patients. Our approach outperforms 6 other state-of-the-art time-series models for both datasets, achieving significantly higher F1 scores ( p <0.05). Ablation studies confirm the effectiveness of both curriculum learning and chaos-aware gating mechanism, highlighting the potential of our framework for clinical prediction in high-noise, low-data neonatal care settings.
- 8 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
Open AccessCurriculum Learning With Chaos-Aware Gating Mechanism for Infant Cardiorespiratory Status Prediction via Heart Rate Variability and Bedside Data
Predicting the cardiorespiratory status of extremely preterm infants using signals from bedside monitors is critical for timely intervention in neonatal intensive care units (NICUs), potentially improving long-term outcomes. However, directly adapting existing time series data analysis methods to this task is challenged by the scarcity of neonatal data and the high noise levels inherent to physiological signals in this population. To address these challenges, we propose a novel hybrid neural network that integrates C urriculum L earning and a C haos-aware G ating mechanism, called CLCG. Given the long multivariate physiological sequences, we first apply a sliding-window strategy to segment the data into overlapping windows. For each window, a shared Convolutional Neural Network (CNN) fuses multi-modal signals and extracts localized representations. To increase data diversity and enhance generalization capability, we introduce a curriculum learning-based data augmentation scheme that progressively reduces the number of windows during training. This encourages the model to learn discriminative representations by exposing it to increasingly challenging scenarios with less temporal context. To further mitigate the impact of signal noise and improve temporal modeling, we extract handcrafted features, the Lyapunov exponent, from the heart rate variability (HRV) parameters. These features are incorporated into a novel chaos-aware gate mechanism within the Long Short-Term Memory (LSTM) network. This mechanism guides the gating process based on nonlinear dynamical properties, and suppresses noise-related fluctuations. Finally, a Multi-Layer Perceptron (MLP) predicts the respiratory and hypotensive status over the next five hours. We evaluate our model on two datasets: an in-house cohort of 55 extremely preterm infants from the Monash Children's Hospital, and the publicly available MC-MED dataset from emergency department patients. Our approach outperforms 6 other state-of-the-art time-series models for both datasets, achieving significantly higher F1 scores ( p <0.05). Ablation studies confirm the effectiveness of both curriculum learning and chaos-aware gating mechanism, highlighting the potential of our framework for clinical prediction in high-noise, low-data neonatal care settings.
- 4 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-18
Advancing In-Context Learning for Efficient and Stable Medical Report Generation
Vision-language models (VLMs) have shown strong generalization across multimodal tasks, but adapting them to medical report generation (MRG) often demands extensive paired image-text data that are limited due to data privacy and annotation cost. In-context learning (ICL) offers a promising training-free alternative, yet standard ICL approaches rely on long demonstration prompts that are computationally inefficient and often yield inconsistent or clinically inaccurate descriptions. To address these challenges, we propose Principal In-Context Vectors (PCVs), a compact latent-guidance framework that distills multimodal demonstrations into stable semantic representations. By extracting hidden states from auto-regressive VLMs and applying principal component analysis (PCA), we identify robust semantic directions that remain stable under input perturbations. These PCVs are then injected into new queries to steer generation toward accurate and clinically meaningful outputs without any model tuning. Extensive experiments on four MRG benchmark datasets show that our approach can enhance both zero-shot and fully supervised generation quality across diverse settings, including cross-center, cross-disease, and longitudinal scenarios. This work provides a lightweight and scalable approach to adapt pre-trained VLMs for practical clinical deployment.
- 4 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-18
Advancing In-Context Learning for Efficient and Stable Medical Report Generation
Vision-language models (VLMs) have shown strong generalization across multimodal tasks, but adapting them to medical report generation (MRG) often demands extensive paired image-text data that are limited due to data privacy and annotation cost. In-context learning (ICL) offers a promising training-free alternative, yet standard ICL approaches rely on long demonstration prompts that are computationally inefficient and often yield inconsistent or clinically inaccurate descriptions. To address these challenges, we propose Principal In-Context Vectors (PCVs), a compact latent-guidance framework that distills multimodal demonstrations into stable semantic representations. By extracting hidden states from auto-regressive VLMs and applying principal component analysis (PCA), we identify robust semantic directions that remain stable under input perturbations. These PCVs are then injected into new queries to steer generation toward accurate and clinically meaningful outputs without any model tuning. Extensive experiments on four MRG benchmark datasets show that our approach can enhance both zero-shot and fully supervised generation quality across diverse settings, including cross-center, cross-disease, and longitudinal scenarios. This work provides a lightweight and scalable approach to adapt pre-trained VLMs for practical clinical deployment.
- 28 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. PP (02780062) , 1
Integrating Anatomical Priors into a Causal Diffusion Model
3D brain MRI studies often examine subtle morphometric differences between cohorts that are hard to detect visually. Given the high cost of MRI acquisition, these studies could greatly benefit from image syntheses, particularly counterfactual image generation, as has been the case for applications in computer vision. However, counterfactual models struggle to produce anatomically plausible MRIs due to a lack of explicit inductive biases to preserve fine-grained anatomical details. This short-coming arises from the training of models that optimize overall image appearance (e.g., via cross-entropy) rather than preserving subtle, yet medically relevant, local variations across subjects. To preserve subtle variations, we propose to explicitly integrate anatomical constraints at the voxel level as priors into a generative diffusion framework. Termed Probabilistic Causal Graph Model (PCGM), the approach captures anatomical constraints via a probabilistic graph module and translates those constraints into spatial binary masks of regions where subtle variations occur. The masks (encoded by a 3D ControlNet) constrain a novel counterfactual denoising UNet, whose encodings are then transferred into high-quality brain MRIs via our 3D diffusion decoder. Extensive experiments across multiple datasets demonstrate that PCGM generates structural brain MRIs of higher quality than several baseline approaches. Furthermore, we show, for the first time, that brain measurements extracted from counterfactuals (generated by PCGM) replicate the subtle effects of a disease on cortical brain regions previously reported in the neuroscience literature. This achievement is an important milestone in the use of synthetic MRIs in studies investigating subtle morphological differences. The codes are available at https://github.com/AndyCA111/PCGM.
- 28 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. PP (02780062) , 1
Integrating Anatomical Priors into a Causal Diffusion Model
3D brain MRI studies often examine subtle morphometric differences between cohorts that are hard to detect visually. Given the high cost of MRI acquisition, these studies could greatly benefit from image syntheses, particularly counterfactual image generation, as has been the case for applications in computer vision. However, counterfactual models struggle to produce anatomically plausible MRIs due to a lack of explicit inductive biases to preserve fine-grained anatomical details. This short-coming arises from the training of models that optimize overall image appearance (e.g., via cross-entropy) rather than preserving subtle, yet medically relevant, local variations across subjects. To preserve subtle variations, we propose to explicitly integrate anatomical constraints at the voxel level as priors into a generative diffusion framework. Termed Probabilistic Causal Graph Model (PCGM), the approach captures anatomical constraints via a probabilistic graph module and translates those constraints into spatial binary masks of regions where subtle variations occur. The masks (encoded by a 3D ControlNet) constrain a novel counterfactual denoising UNet, whose encodings are then transferred into high-quality brain MRIs via our 3D diffusion decoder. Extensive experiments across multiple datasets demonstrate that PCGM generates structural brain MRIs of higher quality than several baseline approaches. Furthermore, we show, for the first time, that brain measurements extracted from counterfactuals (generated by PCGM) replicate the subtle effects of a disease on cortical brain regions previously reported in the neuroscience literature. This achievement is an important milestone in the use of synthetic MRIs in studies investigating subtle morphological differences. The codes are available at https://github.com/AndyCA111/PCGM.
- 21 March 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 120
Artificial intelligence-enhanced Couinaud segmentation for precision liver cancer therapy
- 21 March 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 120
Artificial intelligence-enhanced Couinaud segmentation for precision liver cancer therapy
- Open Access
- 20 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (3) , 1-26
Learning Causality-Aware Exploration with Transformers for Goal-Oriented Navigation
Navigation is a fundamental task in the research of Embodied AI, and recent advances in machine learning algorithms have garnered growing interest in developing versatile Embodied AI systems. However, current research in this domain reveals opportunities for improvement. First, the direct application of RNNs and Transformers often overlooks the distinct characteristics of navigation tasks compared to traditional sequential data modeling. These methods are inherently designed to capture long-term dependencies, which are relatively weak in navigation scenarios, potentially limiting their performance in such tasks. Second, the reliance on task-specific configurations, such as pre-trained modules and dataset-specific logic, compromises the generalizability of these methods. We address these constraints by initially exploring the unique differences between Navigation tasks and other sequential data tasks through the lens of Causality, presenting a causal framework to elucidate the inadequacies of conventional sequential methods for Navigation. By leveraging this causal perspective, we propose Causality-Aware Transformer (CAT) Networks for Navigation, featuring a Causal Understanding Module to enhance the model’s Environmental Understanding capability. Meanwhile, our method is devoid of task-specific inductive biases and can be trained in an End-to-End manner, which enhances the method’s generalizability across various contexts. Empirical evaluations demonstrate that our methodology consistently surpasses benchmark performances across a spectrum of settings, tasks, and simulation environments, specifically, in Object Navigation within RoboTHOR, Objective Navigation, Point Navigation in Habitat, and R2R Navigation. Extensive ablation studies reveal that the performance gains can be attributed to the Causal Understanding Module, which demonstrates effectiveness and efficiency in both Reinforcement Learning and Supervised Learning settings. Additionally, further analysis highlights the robustness of our method, demonstrating its capacity to consistently perform well across diverse experimental settings and varying conditions. This robustness underscores the adaptability and generalizability of our approach, reinforcing its potential for application across a wide range of tasks.
- 20 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (3) , 1-26
Open AccessLearning Causality-Aware Exploration with Transformers for Goal-Oriented Navigation
Navigation is a fundamental task in the research of Embodied AI, and recent advances in machine learning algorithms have garnered growing interest in developing versatile Embodied AI systems. However, current research in this domain reveals opportunities for improvement. First, the direct application of RNNs and Transformers often overlooks the distinct characteristics of navigation tasks compared to traditional sequential data modeling. These methods are inherently designed to capture long-term dependencies, which are relatively weak in navigation scenarios, potentially limiting their performance in such tasks. Second, the reliance on task-specific configurations, such as pre-trained modules and dataset-specific logic, compromises the generalizability of these methods. We address these constraints by initially exploring the unique differences between Navigation tasks and other sequential data tasks through the lens of Causality, presenting a causal framework to elucidate the inadequacies of conventional sequential methods for Navigation. By leveraging this causal perspective, we propose Causality-Aware Transformer (CAT) Networks for Navigation, featuring a Causal Understanding Module to enhance the model’s Environmental Understanding capability. Meanwhile, our method is devoid of task-specific inductive biases and can be trained in an End-to-End manner, which enhances the method’s generalizability across various contexts. Empirical evaluations demonstrate that our methodology consistently surpasses benchmark performances across a spectrum of settings, tasks, and simulation environments, specifically, in Object Navigation within RoboTHOR, Objective Navigation, Point Navigation in Habitat, and R2R Navigation. Extensive ablation studies reveal that the performance gains can be attributed to the Causal Understanding Module, which demonstrates effectiveness and efficiency in both Reinforcement Learning and Supervised Learning settings. Additionally, further analysis highlights the robustness of our method, demonstrating its capacity to consistently perform well across diverse experimental settings and varying conditions. This robustness underscores the adaptability and generalizability of our approach, reinforcing its potential for application across a wide range of tasks.
- Open Access
- 20 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (3) , 1-21
Mitigating Data Redundancy to Revitalize Transformer-Based Long-Term Time Series Forecasting System
Long-term time series forecasting (LTSF) is fundamental to various real-world applications, where Transformer-based models have become the dominant framework due to their ability to capture long-range dependencies. However, these models often experience overfitting due to data redundancy in rolling forecasting settings, limiting their generalization ability particularly evident in longer sequences with highly similar adjacent data. In this work, we introduce CLMFormer, a novel framework that mitigates redundancy through curriculum learning and a memory-driven decoder. Specifically, we progressively introduce Bernoulli noise to the training samples, which effectively breaks the high similarity between adjacent data points. This curriculum-driven noise introduction aids the memory-driven decoder by supplying more diverse and representative training data, enhancing the decoder’s ability to model seasonal tendencies and dependencies in the time series data. To further enhance forecasting accuracy, we introduce a memory-driven decoder. This component enables the model to capture seasonal tendencies and dependencies in the time series data and leverages temporal relationships to facilitate the forecasting process. Extensive experiments on six real-world LTSF benchmarks show that CLMFormer consistently improves Transformer-based models by up to 30%, demonstrating its effectiveness in long-horizon forecasting.
- 20 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (3) , 1-21
Open AccessMitigating Data Redundancy to Revitalize Transformer-Based Long-Term Time Series Forecasting System
Long-term time series forecasting (LTSF) is fundamental to various real-world applications, where Transformer-based models have become the dominant framework due to their ability to capture long-range dependencies. However, these models often experience overfitting due to data redundancy in rolling forecasting settings, limiting their generalization ability particularly evident in longer sequences with highly similar adjacent data. In this work, we introduce CLMFormer, a novel framework that mitigates redundancy through curriculum learning and a memory-driven decoder. Specifically, we progressively introduce Bernoulli noise to the training samples, which effectively breaks the high similarity between adjacent data points. This curriculum-driven noise introduction aids the memory-driven decoder by supplying more diverse and representative training data, enhancing the decoder’s ability to model seasonal tendencies and dependencies in the time series data. To further enhance forecasting accuracy, we introduce a memory-driven decoder. This component enables the model to capture seasonal tendencies and dependencies in the time series data and leverages temporal relationships to facilitate the forecasting process. Extensive experiments on six real-world LTSF benchmarks show that CLMFormer consistently improves Transformer-based models by up to 30%, demonstrating its effectiveness in long-horizon forecasting.
- 18 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (8) , 9050-9063
Distilling Object Detectors via Monte Carlo Dropout
Knowledge distillation (KD) has become a fundamental technique for model compression in object detection tasks. The data noise and training randomness may cause the knowledge of the teacher model to be unreliable, referred to as knowledge uncertainty. Existing methods neglect this uncertainty, potentially hindering the student's capacity to capture and understand latent “dark knowledge”. In this work, we introduce a novel strategy that explicitly incorporates knowledge uncertainty, named Uncertainty-Driven Knowledge Extraction and Transfer (UET). Given the unknown, high-dimensional nature of the knowledge distribution, we employ Monte Carlo dropout to effectively estimate the teacher's uncertainty. Leveraging information theory, we combine uncertainty with deterministic knowledge, enabling the student to benefit from both precision and diversity. UET is a plug-and-play method that integrates seamlessly with existing distillation techniques. We validate our approach through comprehensive experiments across various distillation strategies, detectors, and backbones. Specifically, UET achieves state-of-the-art results, with a ResNet50-based GFL detector obtaining 44.1% mAP on the COCO dataset-surpassing baseline performance by 3.9%.
- 18 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (8) , 9050-9063
Distilling Object Detectors via Monte Carlo Dropout
Knowledge distillation (KD) has become a fundamental technique for model compression in object detection tasks. The data noise and training randomness may cause the knowledge of the teacher model to be unreliable, referred to as knowledge uncertainty. Existing methods neglect this uncertainty, potentially hindering the student's capacity to capture and understand latent “dark knowledge”. In this work, we introduce a novel strategy that explicitly incorporates knowledge uncertainty, named Uncertainty-Driven Knowledge Extraction and Transfer (UET). Given the unknown, high-dimensional nature of the knowledge distribution, we employ Monte Carlo dropout to effectively estimate the teacher's uncertainty. Leveraging information theory, we combine uncertainty with deterministic knowledge, enabling the student to benefit from both precision and diversity. UET is a plug-and-play method that integrates seamlessly with existing distillation techniques. We validate our approach through comprehensive experiments across various distillation strategies, detectors, and backbones. Specifically, UET achieves state-of-the-art results, with a ResNet50-based GFL detector obtaining 44.1% mAP on the COCO dataset-surpassing baseline performance by 3.9%.
- 17 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-15
CoreKD: A Context-Aware Local Region Structural Contrastive Knowledge Distillation Framework for Object Detection
Knowledge distillation (KD) aims to transfer knowledge from a cumbersome teacher to a lightweight student, thereby reducing overall model complexity without sacrificing performance. Current methods tend to focus excessively on pixel-level knowledge transfer while overlooking localized and contextual information. To address this, we propose a novel context-aware local region structural contrastive knowledge distillation framework (CoreKD) for object detection tasks. Specifically, we introduce a patch-based semantic structural distillation (PSD) method, facilitating efficient localized valuable knowledge transfer. This method guides the student model in learning both consistency and diverse local knowledge from the teacher model by integrating the semantic and structural information. We also propose the intra-region contextual (Ita-RC) and inter-region contextual (Ite-RC) constraints for the PSD framework. These constraints could transfer the region-wise contextual information between the student and teacher, significantly enhancing the student detector’s capacity to model contextual dependencies. To validate the efficacy of CoreKD, we conduct extensive experiments on the public MS COCO 2017 and PASCAL VOC datasets. These results demonstrate that CoreKD markedly improves the performance of students in object detection tasks, indicating its ability to transfer valuable knowledge.
- 17 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-15
CoreKD: A Context-Aware Local Region Structural Contrastive Knowledge Distillation Framework for Object Detection
Knowledge distillation (KD) aims to transfer knowledge from a cumbersome teacher to a lightweight student, thereby reducing overall model complexity without sacrificing performance. Current methods tend to focus excessively on pixel-level knowledge transfer while overlooking localized and contextual information. To address this, we propose a novel context-aware local region structural contrastive knowledge distillation framework (CoreKD) for object detection tasks. Specifically, we introduce a patch-based semantic structural distillation (PSD) method, facilitating efficient localized valuable knowledge transfer. This method guides the student model in learning both consistency and diverse local knowledge from the teacher model by integrating the semantic and structural information. We also propose the intra-region contextual (Ita-RC) and inter-region contextual (Ite-RC) constraints for the PSD framework. These constraints could transfer the region-wise contextual information between the student and teacher, significantly enhancing the student detector’s capacity to model contextual dependencies. To validate the efficacy of CoreKD, we conduct extensive experiments on the public MS COCO 2017 and PASCAL VOC datasets. These results demonstrate that CoreKD markedly improves the performance of students in object detection tasks, indicating its ability to transfer valuable knowledge.
- 16 January 2026
- Published by Elsevier in Pattern Recognition
- Vol. 175
Knowledge-guided multi-modality transformer for multi-label genetic mutation prediction
- 16 January 2026
- Published by Elsevier in Pattern Recognition
- Vol. 175
Knowledge-guided multi-modality transformer for multi-label genetic mutation prediction
- Open Access
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (6) , 5595-5603
HC-LLM: Historical-Constrained Large Language Models for Radiology Report Generation
Radiology report generation (RRG) models typically focus on individual exams, often overlooking the integration of historical visual or textual data, which is crucial for patient follow-ups. Traditional methods usually struggle with long sequence dependencies when incorporating historical information, but large language models (LLMs) excel at in-context learning, making them well-suited for analyzing longitudinal medical data. In light of this, we propose a novel Historical-Constrained Large Language Models (HC-LLM) framework for RRG, empowering LLMs with longitudinal report generation capabilities by constraining the consistency and differences between longitudinal images and their corresponding reports. Specifically, our approach extracts both time-shared and time-specific features from longitudinal chest X-rays and diagnostic reports to capture disease progression. Then, we ensure consistent representation by applying intra-modality similarity constraints and aligning various features across modalities with multimodal contrastive and structural constraints. These combined constraints effectively guide the LLMs in generating diagnostic reports that accurately reflect the progression of the disease, achieving state-of-the-art results on the Longitudinal-MIMIC dataset. Notably, our approach performs well even without historical data during testing and can be easily adapted to other multimodal large models, enhancing its versatility.
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (6) , 5595-5603
Open AccessHC-LLM: Historical-Constrained Large Language Models for Radiology Report Generation
Radiology report generation (RRG) models typically focus on individual exams, often overlooking the integration of historical visual or textual data, which is crucial for patient follow-ups. Traditional methods usually struggle with long sequence dependencies when incorporating historical information, but large language models (LLMs) excel at in-context learning, making them well-suited for analyzing longitudinal medical data. In light of this, we propose a novel Historical-Constrained Large Language Models (HC-LLM) framework for RRG, empowering LLMs with longitudinal report generation capabilities by constraining the consistency and differences between longitudinal images and their corresponding reports. Specifically, our approach extracts both time-shared and time-specific features from longitudinal chest X-rays and diagnostic reports to capture disease progression. Then, we ensure consistent representation by applying intra-modality similarity constraints and aligning various features across modalities with multimodal contrastive and structural constraints. These combined constraints effectively guide the LLMs in generating diagnostic reports that accurately reflect the progression of the disease, achieving state-of-the-art results on the Longitudinal-MIMIC dataset. Notably, our approach performs well even without historical data during testing and can be easily adapted to other multimodal large models, enhancing its versatility.