Bing Ji

Multiple Myeloma (MM) is a hematologic malignancy characterized by abnormal plasma cell proliferation and bone destruction. Increasing evidence underscores the immune system’s critical role in MM initiation, therapy and recurrence. Mathematical modeling offers a quantitative framework for analyzing such complex interactions. Existing MM models have provided insights into tumor dynamics but largely overlooked immune regulation. To address this gap, we develop a novel model explicitly incorporating Natural Killer (NK) cells, Cytotoxic T Lymphocytes (CTLs), Regulatory T cells (Tregs) and Cytokines Interleukin-2 (IL-2) and Interleukin-10 (IL-10). The model first integrates two immune evasion mechanisms, antigen escape and an immunosuppressive microenvironment, alongside competition among MM subtypes for limited resources. It also simulates Chimeric Antigen Receptor-T cell (CAR-T) therapy to examine how immunosuppression influences outcomes. Simulations reproduce immune–tumor interactions in the bone microenvironment, reveal that antigen loss exerts a stronger effect on immune escape than immunosuppression alone, and show that increased immune pressure accelerates evolution toward resistant subtypes. Enhancing IL-2 and IL-10 critically modulates CAR-T cell abundance and efficacy. Model predictions agree with experimental data, improving invasion kinetics accuracy over models without immune components. This framework offers a robust basis for studying MM-immune dynamics and advancing immunotherapy modeling.

Cervical spondylotic myelopathy (CSM) and parkinsonian syndromes (PS) present similar motor symptoms, often causing misdiagnosis due to current clinical diagnostic limitations. Misdiagnosis can exacerbate patient conditions or result in unnecessary surgical interventions, thereby increasing surgical risks and the likelihood of serious postoperative complications. This study aims to develop a mixed dual-branch network for classifying CSM patients, PS patients, and healthy individuals using gait data. This study recruits 51 CSM patients, 49 PS patients, and 33 healthy controls. The kinematic data are collected and used to calculate the time series of angle, angular velocity, and angular acceleration for the hip, knee, and ankle joints. From each time series, 20 features are extracted, including the time domain, frequency domain, time-frequency domain, and nonlinear features. A dual-branch model named DCDM-Net is proposed to classify subjects through collaborative decision making (CDM) method, with one branch using ResNet with convolutional block attention module (CBAM) and evidential deep learning (EDL) loss for analyzing time series, and the other employing multilayer perceptron (MLP) for dealing with multi-domain features. DCDM-Net achieves an ACC of 92.35% ± 0.76% and an AUC of 96.70% ± 0.47% in the three-class classification task. Additionally, in binary classification scenarios, the model demonstrates robust performance with an average ACC of 93.13% and AUC of 98.34%. Furthermore, comparative evaluations show that the integrated EDL module surpasses Softmax, MC-Dropout, and Deep Ensembles in uncertainty estimation, yielding the lowest Expected Calibration Error (ECE of 0.0304) and lower Brier score (0.1074), indicating superior reliability. However, cross-dataset OOD validation yielded an AUROC of 0.4022 ± 0.2481 and an AUPR of 0.9699 ± 0.0162, revealing that restricting features to joint angles leads to significant distribution overlap; this conversely validates that angular velocity and acceleration are indispensable for preventing model overconfidence. Interpretable results obtained through the SHapley Additive exPlanations (SHAP) method and the integrated gradients (IG) method are confirmed by clinical findings. Our method provides a promising tool for diagnosing CSM and PS, with the potential to reduce misdiagnosis. The code implementation of this study is available at https://github.com/AImedcinesdu212/DCDM-Net.

The disparity between image and text representations, often referred to as the modality gap, remains a significant obstacle for Vision Language Models (VLMs) in medical image segmentation. This gap complicates multi-modal fusion, thereby restricting segmentation performance. To address this challenge, we propose Evidence-driven Vision Language Model (EviVLM)—a novel paradigm that integrates Evidential Learning (EL) into VLMs to systematically measure and mitigate the modality gap for enhanced multi-modal fusion. To drive this paradigm, an Evidence Affinity Map Generator (EAMG) is proposed to collect complementary cross-modal evidences by learning a global cross-modal affinity map, thus refining modality-specific evidence embedding. An Evidence Differential Similarity Learning (EDSL) is further proposed to collect consistent cross-modal evidences by performing Bias-Variance Decomposition on differential matrix derived from bidirectional similarity matrices between image and text evidence embeddings. Finally, the subjective logic is used for mapping the collected evidences to opinions, and the Dempster-Shafer’s theory based combination rule is introduced for opinion aggregation, thereby quantifying the modality gap and facilitating effective multi-modal integration. Experimental results on three public medical image segmentation datasets validate that the proposed EviVLM can achieve state-of-the-art performance. Code is available at: https://github.com/QingtaoPan/EviVLM.

Low-quality pseudo labels pose a significant obstacle in semi-supervised medical image segmentation (SSMIS), impeding consistency learning on unlabeled data. Leveraging vision-language model (VLM) holds promise in ameliorating pseudo label quality by employing textual prompts to delineate segmentation regions, but it faces the challenge of cross-modal alignment uncertainty due to multiple correspondences (multiple images/texts tend to correspond to one text/image). Existing VLMs address this challenge by modeling semantics as distributions but such distributions lead to semantic degradation. To address these problems, we propose Alignment-Multiplicity Aware Vision-Language Model (AMVLM), a new VLM pre-training paradigm with two novel similarity metric strategies. (i) Cross-modal Similarity Supervision (CSS) proposes a probability distribution transformer to supervise similarity scores across fine-granularity semantics through measuring cross-modal distribution disparities, thus learning cross-modal multiple alignments. (ii) Intra-modal Contrastive Learning (ICL) takes into account the similarity metric of coarse-fine granularity information within each modality to encourage cross-modal semantic consistency. Furthermore, using the pretrained AMVLM, we propose a pioneering text-guided SSMIS network to compensate for the quality deficiencies of pseudo-labels. This network incorporates a text mask generator to produce multimodal supervision information, enhancing pseudo label quality and the model’s consistency learning. Extensive experimentation validates the efficacy of our AMVLM-driven SSMIS, showcasing superior performance across four publicly available datasets. The code will be available at: https://github.com/QingtaoPan/AMVLM.