UngLong/openm3chest-npy-v2
收藏资源简介:
OpenM3Chest NPY v2是一个医学影像数据集,包含约3,793个唯一的胸部CT扫描体积,以NumPy float16格式(.npy文件)存储。数据形状为[Z, H, W],其中Z轴表示切片数量,随扫描不同而变化,通常在100到300片之间。数据以亨斯菲尔德单位(HU)表示,来源于NLST(国家肺癌筛查试验)并通过IDC(影像数据共享平台)获取。该数据集专为MedGemma 1.5-4B-IT模型的LoRA(低秩适应)微调而设计,用于处理胸部CT扫描的医疗影像任务,如结节检测和筛查。数据集还包括预处理步骤,例如使用三个HU窗口(骨/肺、软组织、脑)进行每切片RGB转换,以适配模型输入。
OpenM3Chest NPY v2 is a medical imaging dataset containing approximately 3,793 unique chest CT scan volumes, stored in NumPy float16 format with .npy files. The data shape is [Z, H, W], where the Z-axis represents the number of slices, which varies across scans and typically ranges from 100 to 300 slices. The data is represented in Hounsfield Units (HU), sourced from the National Lung Screening Trial (NLST) and acquired via the Imaging Data Commons (IDC). This dataset is specifically designed for Low-Rank Adaptation (LoRA) fine-tuning of the MedGemma 1.5-4B-IT model to handle medical imaging tasks involving chest CT scans, such as nodule detection and screening. Additionally, the dataset includes preprocessing steps, such as performing per-slice RGB conversion using three HU windows (bone/lung, soft tissue, and brain) to meet the model's input requirements.




