Renal Cyst Measurement

This repository contains code to detect and measure renal cysts on abdominal ultrasound images.

Requirement

For prediction and UNet++ training, this project requires

• cuda8
• python3.6
• numpy
• opencv-python
• matplotlib
• h5py==2.10.0
• Keras==2.2.2
• tensorflow-gpu==1.4.0 #tensorflow-gpu==1.4.1
• scikit-image
• pytorch-lightning
• torch
• torchvision
• pandas
• seaborn

See the YOLOv5 repository for YOLOv5 training requirements.

Data

RenalCystMeasurement/data
├unetpp
│ ├input
│ │ ├image1.png
│ │ ├image2.png
│ │ ├...
│ │
│ └groundTruth
│ │ ├image1.png
│ │ ├image2.png
│ │ ├...
│ 
└yolov5
  ├images
  │ ├image1.png
  │ ├image2.png
  │ ├...
  │
  └labels
    ├image1.txt
    ├image2.txt
    ├...

Format of coordinate.pickle

{'image1.png': ((x1, y1),(x2, y2)),
 'image2.png': [((x1, y1),(x2, y2)),
                ((x3, y3),(x4, y4)),
                ((x5, y5),(x6, y6))],
 ...
 }

Examples

# prediction
python3.6 prediction.py --source img.png

# train YOLOv5
cd yolov5
python3.6 train.py --epoch 100 --data cyst.yaml --weights yolov5m.pt --img 256 --batch 16

# generate heatmap for UNet++
python3.6 generate_heatmap.py

# train UNet++
python3.6 train_unetpp.py

Reference

• YOLOv5
• UNet++