Alshagathrh Fahad, Alzubaidi Mahmood, Alswat Khalid, Aldhebaib Ali, Alahmadi Bushra, Alkubeyyer Meteb, Alosaimi Abdulaziz, Alsadoon Amani, Alkhamash Maram, Schneider Jens, Househ Mowafa
College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar.
Liver Disease Research Center, College of Medicine, King Saud University, Riyadh, Saudi Arabia.
Data Brief. 2024 Dec 27;58:111266. doi: 10.1016/j.dib.2024.111266. eCollection 2025 Feb.
This study presents a comprehensive ultrasound image dataset for Non-Alcoholic Fatty Liver Disease (NAFLD), addressing the critical need for standardized resources in AI-assisted diagnosis. The dataset comprises 10,352 high-resolution ultrasound images from 384 patients collected at King Saud University Medical City and National Guard Health Affairs in Saudi Arabia. Each image is meticulously annotated with NAFLD Activity Score (NAS) fibrosis staging and steatosis grading based on corresponding liver biopsy results. Unlike other datasets that rely on bounding boxes, we opted for full-image labelling based on biopsy findings, which link to histopathological results, ensuring more precise representation of liver conditions. Rigorous pre-processing ensures high-quality image preservation, including expert radiologist assessment, DICOM to PNG conversion, and standardization to 768 × 1024 pixels. This resource supports various computer vision tasks, enabling the development of AI algorithms for accurate NAFLD diagnosis and staging. A large, diverse, and well-annotated dataset like ours is essential for enhancing model performance and generalization, providing a valuable resource for researchers to develop robust AI models in medical imaging.
本研究展示了一个用于非酒精性脂肪性肝病（NAFLD）的全面超声图像数据集，满足了人工智能辅助诊断中对标准化资源的迫切需求。该数据集包含来自沙特阿拉伯国王沙特大学医学城和国民警卫队卫生事务部的384名患者的10352张高分辨率超声图像。每张图像都根据相应的肝活检结果，精心标注了NAFLD活动评分（NAS）、纤维化分期和脂肪变性分级。与其他依赖边界框的数据集不同，我们选择基于活检结果进行全图像标注，这些结果与组织病理学结果相关联，确保更精确地呈现肝脏状况。严格的预处理确保了高质量图像的保存，包括专家放射科医生评估、DICOM到PNG的转换以及标准化为768×1024像素。该资源支持各种计算机视觉任务，有助于开发用于准确NAFLD诊断和分期的人工智能算法。像我们这样一个大型、多样且标注良好的数据集对于提高模型性能和泛化能力至关重要，为研究人员在医学成像中开发强大的人工智能模型提供了宝贵资源。
