kikiru328 / brats20

Brain tumor segmentation with Unet models.

BraTs20

Brain Tumor Segmentation, 두뇌 ct 영상 내 Tumor segmentation을 진행합니다.
알고리즘은 3D Unet, 2D Unet을 적용하여 진행하였습니다.
Challenge Site : BraTs20

BraTs20 Data Info

• Data Format : nii.gz
• Image
  1. T1 : native
  2. T1CE : post_contrast
  3. T2 : T2-weighted
  4. Flair : T2 Fluid Attenuated Inversion Recovery
• Labels
  1. Label 0 : Unlabeled Volumn
  2. Label 1 : Necrotic and non-enhancing tumor core (NCR / NET)
  3. Label 2 : Peritumoral Edema (ED)
  4. Label 3 : missing ( No pixels in all the volumnes contain label 3)
  5. Label 4 : GD-enhancing tumor (ET)

Model

• Performance
  3D UNET : Mean IoU = 0.8143
  2D UNET : Dice_coeff = 1.20 (Multiclass label into Singleclass)

Results

• 3D UNET
  

• 2D UNET