vijula / vnet

Segmentation of Kidney and Tumor Competition A Pytorch implementation for <V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation>

3D Segmentation by VNet

Pipeline

Phase1：Coarse Segmentation for Kidney AND Tumor(consider two organs as one, 0: background, 1: Kidney AND Tumor) to get a bounding box of Kidney and Tumor. Post processing such as Maximum connected domain is applied to remove the noise.

Phase2：Detail Segmentation for Kidney OR Tumor(0: background, 1:Kidney, 2:Tumor).Post processing such as Maximum connected domain, changing threshold in sigmod(times a weight for the tumor probability map)

Model

VNet(Phase 1)

Fig from the paper <V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation>

2-step Model(Phase 2)

Can be easily modified to 2-step model.

Dataset

200 CT cases pre-processed by clip operation to range (-250,250) from KiTS19 competition.

Loss Function

Jointly use

1. CE loss
2. Dice loss
3. Focal loss
   Reference:
   https://blog.csdn.net/m0_37477175/article/details/83004746

Metrics

Dice: Results of Phase1:

Results of Phase2:

boundbox.py

Get a (384,240,80) bounding box from coarse_bbox.txt(generate from the Phase)