2. Material and methods
- ROI
- Tile-based classification
- slide-based classification
2.2 ROI detection with Image Processing
- Using OTSU, multilevel mapping
- First covert the original image from RGB to HSV
- Generate binary mask
- Extract tissue region using tisue mask and image morphology operations.
2.3 Deep-learning CNN for Tile-based Classification
训练三种模型(Inceptionv3, InceptionResnet-v2, Resnet50),然后组合起来变成一个新的模型。
DataAugmentation
2.4 Post-processing
256X256 pixel patch produce heatmap
build classifier for slide-based classification
使用heatmap的特征来进行分类器的训练