Scholarlink

<p>Hereditary Multiple Exostoses (HME) is a rare skeletal disorder characterized by the growth of multiple benign bone tumors. They often lead to orthopedic complications and malignant transformation. Despite advancements in medical imaging, no automated deep learning framework has been specifically developed for detecting HME in hip radiographs. In this study, we propose HME-LiteNet, a lightweight convolutional neural network (CNN) customized to identify HME-related features on pelvic radiographs. To address the challenge of limited data, we performed domain-specific data augmentation to enhance model robustness. The proposed model was benchmarked against a pretrained MobileNetV2, demonstrating its superior performance in terms of computational complexity and inference speed. Furthermore, we conducted acceleration cross-NVIDIA platform training and inference experiments. We used a high-performance GeForce RTX 3050 Graphic Processing Unit (GPU) and an edge-capable Jetson Orin device. The model was further optimized using TensorFlow Lite (TFLite). It achieved an accuracy and area under the receiver operating characteristic curve (AUC-ROC) of 0.99 on the validation set. It also reduces the computational complexity by a factor of 1.64 × compared with MobileNetV2. The inference throughput reached 8.75 GOPS on the RTX3050 and 7.58 GOPS on the Jetson Orin, highlighting its suitability for real-time deployment. Visualization using Grad-CAM confirmed that the model consistently focused on clinically relevant regions in HME-positive radiographs. This approach offers interpretable and reliable prediction. These results demonstrate the potential of AI-assisted diagnosis in enhancing the early detection of HME in both clinical and portable healthcare settings.</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

Edge-deployable AI for a rare disease: automated detection of hereditary multiple exostoses in hip X-rays

  • 期刊论文

  • 发表日期：

    2026年8月3日

查看全文
  • Emna Aridhi,
  • Kaouther Laabidi,
  • Ahlem Maghzaoui,
  • Abdelkader Mami

摘要

Hereditary Multiple Exostoses (HME) is a rare skeletal disorder characterized by the growth of multiple benign bone tumors. They often lead to orthopedic complications and malignant transformation. Despite advancements in medical imaging, no automated deep learning framework has been specifically developed for detecting HME in hip radiographs. In this study, we propose HME-LiteNet, a lightweight convolutional neural network (CNN) customized to identify HME-related features on pelvic radiographs. To address the challenge of limited data, we performed domain-specific data augmentation to enhance model robustness. The proposed model was benchmarked against a pretrained MobileNetV2, demonstrating its superior performance in terms of computational complexity and inference speed. Furthermore, we conducted acceleration cross-NVIDIA platform training and inference experiments. We used a high-performance GeForce RTX 3050 Graphic Processing Unit (GPU) and an edge-capable Jetson Orin device. The model was further optimized using TensorFlow Lite (TFLite). It achieved an accuracy and area under the receiver operating characteristic curve (AUC-ROC) of 0.99 on the validation set. It also reduces the computational complexity by a factor of 1.64 × compared with MobileNetV2. The inference throughput reached 8.75 GOPS on the RTX3050 and 7.58 GOPS on the Jetson Orin, highlighting its suitability for real-time deployment. Visualization using Grad-CAM confirmed that the model consistently focused on clinically relevant regions in HME-positive radiographs. This approach offers interpretable and reliable prediction. These results demonstrate the potential of AI-assisted diagnosis in enhancing the early detection of HME in both clinical and portable healthcare settings.