Multi-Teacher Distillation for Cross-Domain Streaming Electrolaryngeal Speech Encoding
为解决电喉语音编码问题,提出多教师知识蒸馏框架训练轻量级流式内容编码器,通过自监督学习模型和微调识别模型提供目标,降低电喉语音错误率。
为解决电喉语音编码问题,提出多教师知识蒸馏框架训练轻量级流式内容编码器,通过自监督学习模型和微调识别模型提供目标,降低电喉语音错误率。
本文使用结合物理基础的铅笔束剂量引擎与3D卷积残差校正网络的方法,解决了快速准确计算质子剂量的问题。
研究通过病变引导的ROI深度学习方法,提高了卵巢超声分类准确性并减少了标注负担,使用多种模型比较得出MaxViT-Tiny在两个数据集上表现最佳。
This study addresses the challenge of assessing cerebrovascular reserve in Moyamoya disease patients with acetazolamide contraindications by proposing CAE3D, a three-dimensional conditional autoencoder that non-invasively synthesizes post-challenge cerebral blood flow maps from baseline ASL images. This work represents the first validation of retrospectively generating stress perfusion data from baseline MRI using a deterministic architecture integrating perfusion imaging with deep learning. Experimental results demonstrate superior performance over various 3D generative and foundation models, achieving a mean absolute error of 0.066, a structural similarity index of 0.80, and near-zero whole-brain bias. Consequently, this method establishes a novel paradigm for precise hemodynamic assessment in contraindicated patients, effectively bridging advanced deep learning techniques with clinical neuroimaging requirements to overcome diagnostic limitations associated with pharmacological vasodilators.
This work addresses the challenge of nonlinear registration between histological sections and HiP-CT volumetric data, which arises from significant modality discrepancies. To tackle this without requiring paired training data, the authors propose a structure-preserving cross-modal image translation method. Leveraging a frozen DINOv2 backbone as a semantic structural anchor and modality-specific LoRA adapters, the approach enables efficient and generalizable cross-modal representation learning within a cycle-consistent adversarial training framework. This design effectively mitigates content drift while enhancing structural consistency. Experimental results demonstrate that the proposed method outperforms CycleGAN in terms of Fréchet Inception Distance (FID), mutual information, and edge preservation metrics, and substantially improves feature correspondence in downstream registration tasks.
为解决电喉语音编码问题,提出多教师知识蒸馏框架训练轻量级流式内容编码器,通过自监督学习模型和微调识别模型提供目标,降低电喉语音错误率。
本文使用结合物理基础的铅笔束剂量引擎与3D卷积残差校正网络的方法,解决了快速准确计算质子剂量的问题。
研究通过病变引导的ROI深度学习方法,提高了卵巢超声分类准确性并减少了标注负担,使用多种模型比较得出MaxViT-Tiny在两个数据集上表现最佳。
This study addresses the challenge of assessing cerebrovascular reserve in Moyamoya disease patients with acetazolamide contraindications by proposing CAE3D, a three-dimensional conditional autoencoder that non-invasively synthesizes post-challenge cerebral blood flow maps from baseline ASL images. This work represents the first validation of retrospectively generating stress perfusion data from baseline MRI using a deterministic architecture integrating perfusion imaging with deep learning. Experimental results demonstrate superior performance over various 3D generative and foundation models, achieving a mean absolute error of 0.066, a structural similarity index of 0.80, and near-zero whole-brain bias. Consequently, this method establishes a novel paradigm for precise hemodynamic assessment in contraindicated patients, effectively bridging advanced deep learning techniques with clinical neuroimaging requirements to overcome diagnostic limitations associated with pharmacological vasodilators.
This work addresses the challenge of nonlinear registration between histological sections and HiP-CT volumetric data, which arises from significant modality discrepancies. To tackle this without requiring paired training data, the authors propose a structure-preserving cross-modal image translation method. Leveraging a frozen DINOv2 backbone as a semantic structural anchor and modality-specific LoRA adapters, the approach enables efficient and generalizable cross-modal representation learning within a cycle-consistent adversarial training framework. This design effectively mitigates content drift while enhancing structural consistency. Experimental results demonstrate that the proposed method outperforms CycleGAN in terms of Fréchet Inception Distance (FID), mutual information, and edge preservation metrics, and substantially improves feature correspondence in downstream registration tasks.