Keras with mat for U-net

Implementation of the Keras U-Net for ".mat" file of MATLAB.

Image pattern extraction using U-Net

---

Requirements

(Windows) Python 3.5.2 ver.

Details are shown below.

Keras==2.0.4

tensorflow-gpu==1.6.0

scipy==1.0.0

numpy==1.14.2

matplolib==2.1.1

---

Augmentation

image augmentation with rotation 90 degree, up-down filp, left-right flip randomly.

---

Model

U-Net-based model for pattern extraction

U-Net

U-Net: Convolutional Networks for Biomedical Image Segmentation https://arxiv.org/abs/1505.04597

Reference code

https://github.com/jocicmarko/ultrasound-nerve-segmentation/

---

Results

Deep-Learning results

A few slices of Input image, Label image, Result image comparison

(a) Input image (with field inhomogeneity artifact in MRI, out-of-phase angle image)

(b) Label image (with field inhomogeneity artifact removal "SUPER" method in MRI)

"SUPER" method https://synapse.koreamed.org/DOIx.php?id=10.13104/imri.2018.22.1.37

(c) Deep-Learning result image

---

Water-Fat seperation result

(a) Before artifact removal

(b) After artifact removal with "SUPER" method

(c) After artifact removal with trained "SUPER" method Deep-Learning

---

Feature visualization

(a) Conv2D layer1

(b) Conv2D layer2

(c) Conv2D layer3

(d) Conv2D layer4

(e) Conv2D layer5

(f) Deconvolution layer1

(g) Deconvolution layer2

(h) Deconvolution layer3

(i) Deconvolution layer4