3CB Software and Neural Network

This repo contains the three compartment breast (3CB) final trained neural network model used in the manuscript entitled "Dual-energy three compartment breast imaging (3CB) for novel compositional biomarkers to improve detection of malignant lesions", submitted to Nature publications in q1 2021.

The neural network model takes 3CB derived features and CAD predicted probability of malignancy and outputs a new probability of malignancy.

---

Requirements

Hardware requirements

Minimum computational hardware

• 4 core CPU

• 8 gb of RAM

• Cuda enabled GPU (optional)

  Runtimes:

  • neural network prediction: < 10 seconds per 1000 ROIs
  • demo runtime: < 1 minute

Neural Network Model

• Python 3.6+, with recent versions of the following python packages:
  • cv2
  • matplotlib
  • numpy
  • pandas
  • scipy
  • seaborn
  • Keras
  • Tensorflow

DICOM viewer

• Fiji/Imagej (optional)

---

Repository folders

Neural Network

• nn_model: final trained model referenced in manuscript
  • *tracked via git-lfs

Demo

• example data and driver program to extract 3CB features and run NN predictions

---

Neural Network Run Overview

# load scaler
sc=load('path to scaler *.bin')

# load and use model
from keras.models import Sequential, load_model
model=load_model("path to model *.h5")

---

Contact

• Lambert Leong: lambert3@hawaii.edu
• Thomas Wolfgruber: tomwolf@hawaii.edu
• John Shepherd: johnshep@hawaii.edu

---