888 low-dose CT scans containing 1,186 annotated lung nodules verified by multiple radiologists. The data includes spatial coordinates for nodule candidates and diameter measurements for confirmed lesions across 10 subsets.
Use Cases
- Train 3D convolutional neural networks to localize nodules using the 'coordX', 'coordY', and 'coordZ' spatial coordinates.
- Perform binary classification for false positive reduction using the 'class' column in the candidates list.
- Benchmark volumetric segmentation algorithms against the 'diameter_mm' ground truth values.
Strengths
- 888 CT scans provided in MetaImage (.mhd) format with corresponding raw pixel data.
- 1,186 nodules annotated with 'coordX', 'coordY', 'coordZ', and 'diameter_mm' attributes.
- 551,065 candidate locations provided in 'candidates.csv' for training false positive reduction models.
- Annotations require agreement from at least three out of four expert radiologists to ensure high label quality.