Scholarlink

<p>Cockayne Syndrome (CS) is an autosomal recessive rare neurodegenerative disorder with features of growth failure, microcephaly, and premature aging. It is challenging to diagnose and classify a subtype, particularly to distinguish Type I, II, and III due to the scarcity of datasets and overlapping phenotypes. This paper introduces a novel deep learning framework, the Parallel Capsule ConvNet Attention Architecture, which integrates Capsule-Mobius Convolutional Routing (CMCR) with Honey Badger Optimization (HBO) for improved diagnostic performance. The architecture utilizes parallel Capsule ConvNet architecture that simulates human cognitive vision by preserving spatial hierarchies between symptoms and clinical markers. CMCR features a biologically inspired Möbius-type capsule routing system facilitating unbroken contextual flow and topological feature preservation between layers. This enables the model to leverage subtle morphological clues essential for subtype differentiation. To encourage convergence and prevent suboptimal learning, HBO adaptively varies weights, dropout rates, and attention parameters during training through swarm-based behavioral modes. Tested on a preprocessed multi-modal dataset of MRI scans, facial images, and clinical histories of confirmed CS patients, the developed model demonstrates high diagnostic accuracy. It achieves an F1-score of 94.6%, sensitivity of 92.8%, and displays improved interpretability through attention heat maps.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Parallel capsule ConvNet attention architecture for diagnosis and classification of Cockayne syndrome

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • K. Malathi

摘要

Cockayne Syndrome (CS) is an autosomal recessive rare neurodegenerative disorder with features of growth failure, microcephaly, and premature aging. It is challenging to diagnose and classify a subtype, particularly to distinguish Type I, II, and III due to the scarcity of datasets and overlapping phenotypes. This paper introduces a novel deep learning framework, the Parallel Capsule ConvNet Attention Architecture, which integrates Capsule-Mobius Convolutional Routing (CMCR) with Honey Badger Optimization (HBO) for improved diagnostic performance. The architecture utilizes parallel Capsule ConvNet architecture that simulates human cognitive vision by preserving spatial hierarchies between symptoms and clinical markers. CMCR features a biologically inspired Möbius-type capsule routing system facilitating unbroken contextual flow and topological feature preservation between layers. This enables the model to leverage subtle morphological clues essential for subtype differentiation. To encourage convergence and prevent suboptimal learning, HBO adaptively varies weights, dropout rates, and attention parameters during training through swarm-based behavioral modes. Tested on a preprocessed multi-modal dataset of MRI scans, facial images, and clinical histories of confirmed CS patients, the developed model demonstrates high diagnostic accuracy. It achieves an F1-score of 94.6%, sensitivity of 92.8%, and displays improved interpretability through attention heat maps.