Data Augmentation-driven Few-shot Scribble Supervised Medical Image Segmentation

Introduction

Data Augmentation-driven Few-shot Scribble Supervised Medical Image Segmentation.

Methods

• Dataset: The MS-CMRSeg 2019 Challenge, LGE modality from 45 cardiomyopathy patients.
• Augmentation: Mixup, Cutmix, Cutout, PuzzleMix, Comixup.
• Setup: U-Net.
• Evaluation: Dice Score.

Results

Visualization of scribble:

Dice score performance on MSCMRseg dataset:

Visualization of segmentation: