University of Wisconsin Milwaukee

UWM Digital Commons

Theses and Dissertations

May 2020

Medical Image Segmentation with Deep Learning

Chuanbo Wang
University of Wisconsin-Milwaukee

Follow this and additional works at: https://dc.uwm.edu/etd

Part of the Electrical and Electronics Commons

Recommended Citation
Wang, Chuanbo, "Medical Image Segmentation with Deep Learning" (2020). Theses and Dissertations.
2434.
https://dc.uwm.edu/etd/2434

This Thesis is brought to you for free and open access by UWM Digital Commons. It has been accepted for
inclusion in Theses and Dissertations by an authorized administrator of UWM Digital Commons. For more
information, please contact open-access@uwm.edu.
MEDICAL IMAGE SEGMENTATION WITH DEEP LEARNING
by

Chuanbo Wang

A Thesis Submitted in

Partial Fulfillment of the

Requirements for the Degree of

Master of Science

in Engineering

at

The University of Wisconsin-Milwaukee

May 2020
 ABSTRACT

MEDICAL IMAGE SEGMENTATION WITH DEEP LEARNING

by

Chuanbo Wang

The University of Wisconsin-Milwaukee, 2016

Under the Supervision of Zeyun Yu

Medical imaging is the technique and process of creating visual representations

of the body of a patient for clinical analysis and medical intervention. Healthcare

professionals rely heavily on medical images and image documentation for proper

diagnosis and treatment. However, manual interpretation and analysis of medical

images is time-consuming, and inaccurate when the interpreter is not well-trained.

Fully automatic segmentation of the region of interest from medical images have

been researched for years to enhance the efficiency and accuracy of understanding

such images. With the advance of deep learning, various neural network models

have gained great success in semantic segmentation and spark research interests in

medical image segmentation using deep learning. We propose two convolutional

frameworks to segment tissues from different types of medical images.

Comprehensive experiments and analyses are conducted on various segmentation

neural networks to demonstrate the effectiveness of our methods. Furthermore,

datasets built for training our networks and full implementations are published.

ii
© Copyright by Chuanbo Wang, 2020
All Rights Reserved

iii
 TABLE OF CONTENTS

TABLE OF CONTENTS ..........................................................................................iv

LIST OF FIGURES .................................................................................................vi

LIST OF TABLES ................................................................................................... vii

Chapter 1: Introduction ........................................................................................... 1

Chapter 2: Dataset.................................................................................................. 8

2.1 The Wound Dataset ................................................................................... 8

2.1.1 Dataset Construction ....................................................................... 8

2.1.2 Data Annotation ............................................................................... 9

2.2 The IVD Dataset ...................................................................................... 10

Chapter 3: Methods .............................................................................................. 11

3.1 Wound Segmentation .............................................................................. 11

3.1.1 Model Architecture Overview ......................................................... 11

3.1.2 Transfer Learning .......................................................................... 13

3.1.3 Post-processing ............................................................................. 13

3.2 IVD Segmentation ................................................................................... 15

3.2.1 Localization Network ..................................................................... 15

3.2.2 Segmentation Network .................................................................. 16

3.2.3 Post-processing ............................................................................. 17

Chapter 4: Results ................................................................................................ 18

4.1 Wound Segmentation Results ................................................................. 18

iv
 4.1.1 2D Evaluation Metrics.................................................................... 18

4.1.2 Experiment setup........................................................................... 19

4.1.3 Comparing our method to the others ............................................. 20

4.1.4 Comparison within the Medetec Dataset ....................................... 22

4.2: IVD Segmentation Results...................................................................... 23

4.2.1 3D Evaluation Metrics.................................................................... 23

4.2.2 Effectiveness of the multi-modality data ........................................ 23

4.2.3 Comparison of our methods and state-of-the-art methods ............ 25

Chapter 5: Conclusions ........................................................................................ 26

References ........................................................................................................... 27

v
 LIST OF FIGURES

Figure 1. An illustration of images in our dataset. ............................................... 8

Figure 2. The encoder-decoder architecture of MobilenetV2............................ 10

Figure 3. A demonstration of a depth-separable convolution layer compared to

an original convolution layer ............................................................... 12

Figure 4. An illustration of the segmentation result and the post-processing

method.. .............................................................................................. 14

Figure 5. An overview of the model architecture of the 3D segmentation network

............................................................................................................ 16

vi
 LIST OF TABLES

Table 1. Evaluation on our dataset. .................................................................. 20

Table 2. Comparison of total numbers of trainable parameters. ....................... 22

Table 3. Evaluation on the Medetec dataset. .................................................... 23

Table 4. Segmentation performance of our 3D method using different

combinations of modalities as the training dataset ..................................... 25

Table 5. Segmentation result evaluation of the conventional 3D U-Net, UNICHK,

UNIJLU and our method. ........................................................................... 26

vii
 Chapter 1: Introduction

For accurate diagnosis and proper treatment planning, healthcare professionals

rely heavily on medical images, including computed tomography (CT) images,

magnetic resonance imaging (MRI) scans, and natural images taken in clinical settings.

Such images are further measured and analyzed to provide quantitative parameters

in the diagnosis and treatment. Traditionally, this process is performed manually by

specialists. However, this process is tedious and time-consuming given the large

number of images involved for each patient. Furthermore, the shortage of medical

resources and clinicians in primary and rural healthcare settings decreases the access

and quality of care to millions of Americans. Consequently, research interests in

automatic segmentation and measurement from medical images were captured,

especially in the fields of intervertebral discs segmentation from 3D MRI scans and

wound segmentation from 2D images. Such studies can be roughly categorized into

two groups: traditional methods and deep learning methods.

Studies in the first group focus on combining computer vision techniques and

traditional machine learning approaches. These studies apply manually designed

feature extraction to build a dataset that is later used to support machine learning

algorithms. [1] proposed an algorithm to segment the wound region from 2D images.

49 features are extracted from a wound image using K-means clustering, edge

detection, thresholding, and region growing in both grayscale and RGB. These

features are filtered and prepared into a feature vector that is used to train a Multi-

1
Layer Perceptron (MLP) and a Radial Basis Function (RBF) neural network to identify

the region of a chronic wound. [2] proposed an intervertebral discs (IVD) segmentation

method applied to chest MRI scans. The method solves an energy minimization

problem by graph-cuts algorithms where the graph edges are divided into two types:

terminal edges that connect the voxels and non-terminal edges that connect neighbor

voxels. [3] proposed to generate a Red-Yellow-Black-White (RYKW) probability map

of an input image with a modified hue-saturation-value (HSV) model. This map then

guides the region of interest (ROI) segmentation process using either optimal

thresholding or region growing. [4] and [5] applied an atlas-based method that first

proposes atlas candidates as initialization and then utilize label fusion to combine IVD

atlases to generate the segmentation mask. However, to generate the initialization, [4]

registers IVD atlases to the localization obtained by integral channel features and a

graphical parts model. Whereas [5] uses data-driven regression to create a probability

map, which further defines an ROI as the initialization for segmentation. [6]

demonstrated a wound segmentation method using an energy-minimizing discrete

dynamic contour algorithm applied to the saturation plane of the image in its HSV color

model. The wound area is then calculated from a flood fill inside the enclosed contour.

Another regression-based IVD segmentation method [7] was proposed to address the

segmentation of multiple anatomic structures in multiple anatomical planes from

multiple imaging modalities with a sparse kernel machines-based regression. A 2D

segmentation method proposed in [8] applied an Independent Component Analysis

2
(ICA) algorithm to the pre-processed RGB images to generate hemoglobin-based

images, which are used as input of K-means clustering to segment the granulation

tissue from the wound images. These segmented areas are utilized as an assessment

of the early stages of ulcer healing by detecting the growth of granulation tissue on

ulcer bed. [9] proposed a similar system to segment the burn wounds from 2D images.

Cr-Transformation and Luv-Transformation are applied to the input images to remove

the background and highlight the wound region. The transformed images are

segmented with a pixel-wise Fuzzy C-mean Clustering (FCM) algorithm. [10] proposes

an automatic method using a conditional random field (CRF) based on super-voxels

generated from a variant of simple linear iterative clustering (SLIC). A support vector

machine (SVM) is then used to perform super-voxels classification, which is later

integrated into the potential function of the CRF for final segmentation using graph

cuts. [11] builds an automatic IVD segmentation framework that localizes the vertebral

bodies using regression-forests-based landmark localization and optimizes the

landmarks by a high-level Markov Random Field (MRF) model of global configurations.

The IVD segmentation mask is then generated from an image processing pipeline that

optimizes the convex geodesic active contour based on the geometrical similarity to

IVDs. In [12], IVD segmentation is performed by iteratively deforming the

corresponding average disc model towards the edge of each IVD, in which edge

voxels are defined by a 26-dimension feature vector including intensity, gradient

orientation and magnitude, self-similarity context (SSC) descriptor, and Canny edge

3
descriptor, etc. This group of methods suffers from at least one of the following

limitations: 1) As in many traditional computer vision systems, the computation

complexity is high in the segmentation pipeline, 2) They depend on manually tuned

parameters and empirically handcrafted features which does not guarantee an optimal

result. Additionally, they are not immune to severe pathologies and rare cases, which

are very impractical from a clinical perspective, and 3) The performance is evaluated

on a small biased dataset.

Since the achievements AlexNet [13] accomplished in the ImageNet large scale

visual recognition challenge [14] in 2012, the success of deep learning in the domain

of computer vision sparked interests in semantic segmentation [15] using deep

convolutional neural networks (CNN) [16]. Traditional computer vision and machine

learning methods typically make decisions based on feature extraction. Thus, to find

the segmentation mask, one must guess which wound features are important and then

design sophisticated algorithms that capture these features. However, in CNN, feature

extraction and decision making are integrated. The features are extracted by

convolutional kernels and their importance is determined by the network during the

training process. A typical CNN architecture consists of convolutional layers and a fully

connected layer as the output layer, which requires fixed-size inputs. One successful

variant of CNN is fully convolutional neural networks (FCN) [17]. Networks of this type

are composed of convolutional layers without any fully connected layer at the end of

the network. This enables the network to take arbitrary input sizes and prevent the

4
loss of spatial information caused by the fully connected layers in CNNs. Several FCN-

based methods have been proposed to solve the wound segmentation problem. [18]

estimated the wound area by segmenting wounds with the vanilla FCN architecture

[17]. With time-series data consisting of the estimated wound areas and corresponding

images, wound healing progress is predicted using a Gaussian process regression

function model. However, the mean Dice accuracy of the segmentation is only

evaluated to be 64.2%. [19] proposed to employ the FCN-16 architecture on the wound

images in a pixel-wise manner that each pixel of an image is predicted to which class

it belongs. The segmentation result is simply derived from the pixels classified as a

wound. By testing different FCN architectures, they are able to achieve a Dice

coefficient of 79.4% on their dataset. However, the network’s segmentation accuracy

is limited in distinguishing small wounds and wounds with irregular borders as the

tendency is to draw smooth contours. [20] proposed a new FCN architecture that

replaces the decoder of the vanilla FCN with a skip-layer concatenation up-sampled

with bilinear interpolation. A pixel-wise softmax layer is appended to the end of the

network to produce a probability map, which is post-processed to be the final

segmentation. A dice accuracy of 91.6% is achieved on their dataset with 950 images

taken under an uncontrolled lighting environment with a complex background.

However, images in their dataset are semi-automatically annotated using a watershed

algorithm. This means that the deep learning model is learning how the watershed

algorithm labels wounds as opposed to human specialists.

5
 FCNs are also adopted to solve the IVD segmentation problem. [21] extends the

2D FCN into a 3D version with end-to-end learning and inference. [22] proposes a 3D

multi-scale FCN that expands the typical single-path FCN to three pathways where

each pathway takes volumetric regions on a different scale. Features from three

pathways are then concatenated to generate a probability map, from which the final

3D segmentation mask is generated by simple thresholding. More recently, a modified

FCN, U-Net [45] and its variants have outperformed the state-of-art in many

biomedical image segmentation tasks. The pertinacious architecture and affluent data

augmentation allow U-Net to quickly converge to the optimal model from a limited

number of annotated samples. Comparing to CNN and vanilla FCN, U-Net uses skip

connections between contraction and expansion and a concatenation operator instead

of a sum, which could provide more local information to global information while

expansion. Moreover, U-Net is symmetric such that feature maps in an expansive path

facilitate to transfer more information. U-Net has been widely applied to the IVD

segmentation problem. [23] applies a conventional 3D U-Net [24] on the IVD dataset

provided by the 3rd MICCAI Challenge [25] of Intervertebral Discs Localization and

Segmentation. [26] designs a new network architecture based on U-Net, boundary

specific U-Net (BSU). The architecture consists of repeated application of BSU pooling

layers and residual blocks, following the idea of residual neural networks (RNN). [27]

extends the conventional U-Net by adding three identical pathways in the contracting

path to process the multi-modality channels of the input. These pathways are

6
interconnected with hyper-dense connections to better model relationships between

different modalities in the multi-modal input images. [28] proposes an IVD

segmentation pipeline that first segments the vertebral bodies using a conventional

2D U-Net to find the spine curve and IVD centers. Transverse 2D images and sagittal

3D patches are cropped around the centers to train an RNN fusing both 2D and 3D

convolutions. However, the effectiveness of data augmentation and multi-modality

input images are not fully explored in these works.

To better explore the capacity of deep learning on the wound segmentation and

IVD segmentation problem, we propose two frameworks to automatically segment ROI

from medical images. The first framework is built above a 2D network, MobileNetsV2

[29], to tackle the wound segmentation problem. This network is light-weight and

computationally efficient since significantly fewer parameters are used during the

training process. We built a large dataset of wound images with segmentation

annotations done by wound specialists. This is by far the largest dataset focused on

wound segmentation to the best of our knowledge. The second framework proposed

is built upon a 3D network, 3D U-Net [24], to segment IVD from MRI scans. We

adopted a two-stage pipeline: localizing the IVDs followed by segmenting IVDs based

on the localization. To examine the effectiveness of different combinations of

modalities, various modalities are analyzed with respect to image properties of the

input data based on our analysis in the conducted experiments.

7
Figure 1. An illustration of images in our dataset. The first row contains the raw images collected. The
second row consists of segmentation mask annotations we create with the AZH wound and vascular
center.

Chapter 2: Dataset

2.1 The Wound Dataset

2.1.1 Dataset Construction

There is currently no public dataset large enough for training deep-learning-based

models for wound segmentation. To explore the effectiveness of wound segmentation

using deep learning models, we collaborated with the Advancing the Zenith of

Healthcare (AZH) Wound and Vascular Center, Milwaukee, WI. Our chronic wound

dataset was collected over 2 years at the center and includes 1,109 foot ulcer images

taken from 889 patients during multiple clinical visits. The raw images were taken by

digital single-lens reflex cameras and iPads under uncontrolled illumination conditions,

with various backgrounds. Fig. 1 shows some sample images in our dataset.

The raw images collected are of various sizes and cannot be fed into our deep

learning model directly since our model requires fixed-size input images. To unify the

size of images in our dataset, we first localize the wound by placing bounding boxes

around the wound using an object localization model we trained de novo, YOLOv3

[31]. Our localization dataset contains 1,010 images, which are also collected from the

AZH Wound and Vascular Center. We augmented the images and built a training set

8
containing 3645 images and a testing set containing 405 images. For training our

model we have used LabelImg [32] to manually label all the data (both for training and

testing). The YOLO format has been used for image labeling. The model has been

trained with a batch size of 8 for 273 epochs. With an intersection over union (IoU)

rate of 0.5 and non-maximum suppression of 1.00, we get the mean Average Precision

(mAP) value of 0.939. In the next step, image patches are cropped based on the

bounding boxes result from the localization model. We unify the image size (224 pixels

by 224 pixels) by applying zero-padding to these images, which are regarded in our

dataset data points.

2.1.2 Data Annotation

During training, a deep learning model is learning the annotations of the training

dataset. Thus, the quality of annotations is essential. Automatic annotation generated

with computer vision algorithms is not ideal when deep learning models are trained to

learn how human experts recognize the wound region. In our dataset, the images were

manually annotated with segmentation masks that were further reviewed and verified

by wound care specialists from the collaborating wound clinic. Initially only foot ulcer

images were annotated and included in the dataset as these wounds tend to be

smaller than other types of chronic wounds, which makes it easier and less time-

consuming to manually annotate the pixel-wise segmentation masks. In the future we

plan to create larger image libraries to include all types of chronic wounds, such as

venous leg ulcers, pressure ulcers, and surgery wounds as well as non-wound

reference images. The AZH Wound and Vascular Center, Milwaukee, WI, had

consented to make our dataset publicly available.

9
 2.2 The IVD Dataset

The IVD dataset [30], by courtesy of Prof. Guoyan Zheng from the University of

Bern, consists of 8 sets of 3D multi-modality MRI spine images collected from 8

patients in 2 different stages of prolonged bed test. Each spine image contains at least

7 IVDs of the lower spine (T1-L5) and four modalities following Dixon protocol: in-

phase (inn), opposed-phase (opp), fat and water (wat) images. In detail, water images

are spin echo images acquired from water signals. fat images are spin echo images

acquired from water signals. In-phase images are the sum of water images and fat

images. Opposed-phase images are the difference between water images and fat

images. In total, there are 32 3D single-modality volumes and 66 IVDs. For each IVD,

segmentation ground truth is composed of binary masks manually labeled by three

trained raters under the guidance of clinicians.

Figure 2. The encoder-decoder architecture of MobilenetV2.

10
 Chapter 3: Methods

3.1 Wound Segmentation

In this section we describe our method with the architecture of the deep learning

model for wound segmentation. The transfer learning used during the training of our

model and the post-processing methods including hole filling and removal of small

noises are also described.

3.1.1 Model Architecture Overview

A convolutional neural network (CNN), MobileNetV2 [29], is adopted to segment

the wound from the images. Compared with conventional CNNs, this network

substitutes the fundamental convolutional layers with depth-wise separable

convolutional layers [33] where each layer can be separated into a depth-wise

convolution layer and a point-wise convolution layer. A depth-wise convolution

performs lightweight filtering by applying a convolutional filter per input channel. A

point-wise convolution is a 1 × 1 convolution responsible for building new features

through linear combinations of the input channels. This substitution reduces the

computational cost compared to traditional convolution layers by almost a factor of k2

where k is the convolutional kernel size. Thus, depth-wise separable convolutions are

much more computationally efficient than conventional convolutions suitable for

mobile or embedded applications where computing resource is limited. For example,

the mobility of MobileNetV2 could benefit medical professionals and patients by

11
 (a) (b)

(c)
Figure 3. (a) A depth-separable convolution block. The block contains a 3 × 3 depth-wise
convolutional layer and a 1 × 1 point-wise convolution layer. Each convolutional layer is followed by
batch normalization and Relu6 activation. (b) An example of a convolution layer with a 3 × 3 × 3
kernel. (c) An example of a depth-wise separable convolution layer equivalent to (b).

allowing instant wound segmentation and wound area measurement immediately after

the photo is taken using mobile devices like smartphones and tablets. An example of

a depth-wise separable convolution layer is shown in Figure 3(c), compared to a

traditional convolutional layer shown in Figure 3(b).

The model has an encoder-decoder architecture as shown in Figure 2. The

encoder is built by repeatedly applying the depth-separable convolution block (marked

with diagonal lines in Figure 2). Each block, illustrated in Figure 3(a), consists of six

layers: a 3 × 3 depth-wise convolutional layer followed by batch normalization and

rectified linear unit (Relu) activation [34], and a 1 × 1 point-wise convolution layer

followed again by batch normalization and Relu activation. To be more specific, Relu6

12
[35] was used as the activation function. In the decoder, shown in Figure 2, the

encoded features are captured in multiscale with a spatial pyramid pooling block, and

then concatenated with higher-level features generated from a pooling layer and a

bilinear up-sampling layer. After the concatenation, we apply a few 3 × 3 convolutions

to refine the features followed by another simple bilinear up-sampling by a factor of 4

to generate the final output. A batch normalization layer is inserted into each bottleneck

block and a dropout layer is inserted right before the output layer. In MobileNetV2, a

width multiplier α is introduced to deal with various dimensions of input images. we let

α = 1 thus the input image size is set to 224 pixels × 224 pixels in our model

3.1.2 Transfer Learning

To make the training more efficient, we used transfer learning for our deep learning

model. Instead of randomly initializing the weights in our model, the MobileNetV2

model, pre-trained on the Pascal VOC segmentation dataset [36] is loaded before the

model is trained on our dataset. Transfer learning with the pre-trained model is

beneficial to the training process in the sense that the weights converge faster and

better.

3.1.3 Post-processing

Post Processing, including hole filling and removal of small regions, is performed

to improve the segmentation results as shown in Figure 4. We notice that abnormal

tissue like fibrinous tissue within chronic wounds could be identified as non-wound and

13
cause holes in the segmented wound regions. Such holes are detected by finding

small connected components in the segmentation results and filled to improve the true

positive rate using connected component labeling (CCL) [37]. The small noises are

removed in the same way. The images in the dataset are cropped from the raw image

for each wound. So, we simply remove noises in the results by removing the connected

component small enough based on adaptive thresholds. To be more specific, a

connected region is removed when the number of pixels within the region is less than

a threshold, which is adaptively calculated based on the total number of pixels

segmented as wound pixels in the image.

Figure 4. An illustration of the segmentation result and the post processing method. The first row illustrates
images in the testing dataset. The second row shows the segmentation results predicted by our model
without any post processing. The holes are marked with red boxes and the noises are marked with yellow
boxes. The third row shows the final segmentation masks generated by the post processing method.

14
 3.2 IVD Segmentation

In our proposed 3D method, a two-stage coarse-to-fine strategy is used to tackle

the segmentation problem directly on 3D volumes. The general workflow is illustrated

in Fig. 3. In the first stage, each IVD is localized and a voxel is assigned as its center.

These centers are used to divide the volume into small 3D patches, each of which

contains a single IVD. In the second stage, a multimodal deep learning model is trained

on the patches for precise segmentation.

Medical images are often complex and noisy in nature where ROI is relatively

small comparing to the background. We first localize the IVDs in the image and then

crop 3D patches based on the localization. This not only gets rid of some background

but simplifies the problem for the segmentation stage and reduces the computational

cost as well. It has been shown that 3D U-Net achieves the best localization result but

not the best segmentation result [25] We use this two-stage strategy to work around

this problem. In the end, post-processing is performed to generate the final

segmentation.

3.2.1 Localization Network

For the localization of IVDs, we train a localization network, which is a conventional

3D U-net, on the IVD dataset to roughly locate the IVDs from the volume. From this

segmentation, we have a good estimate of IVD centers by finding the center of each

connected component after removing small regions.

15
 Figure 5. An overview of the model architecture of the 3D segmentation network

From our observation, IVDs are generally sparsely located in 3D space with a

distance from each other and share a common disc-like morphological profile. Thus,

we simply put a 35*35*25 bounding box around each estimated center to crop a 3D

patch. Then we zero-pad the patches to 36*36*28 so they can be nicely fed into the

segmentation network in the next stage described below.

3.2.2 Segmentation Network

For IVD segmentation from the 3D patches, we employ a modified 3D U-Net

architecture that essentially looks at IVD segmentation as a regression problem. This

network takes 3D patches as input and predicts 3D patches where the intensity value

on each voxel stands for how confident is the network in the voxel belonging to an IVD.

Figure 5 presents an overview of the architecture of our 3D segmentation network.

Each step in the contracting path consists of repeated application of two 3x3x3

unpadded 3D convolutions followed by a Relu. A dropout operation is inserted between

16
the two convolutions to reduce the dependence on the training dataset and increase

the accuracy. A dropout rate of 0.2 is used following the analysis [38] on the dropout

effect in CNN. We also apply batch normalization to speed up and stabilize the training

process and a 2x2x2 max pooling layer with stride 2 for down-sampling after every two

convolutional layers. At each down-sampling step, we double the number of feature

channels. Every step in the expansive path consists of an up-sampling of the feature

map followed by a 2x2x2 up convolution that halves the number of feature channels,

a concatenation with the corresponding feature map from the contracting path, and

two 3x3x3 convolutions, each followed by a Relu. The output layer is a 1x1x1

convolution layer with sigmoid activation used to generate the segmentation mask for

each modality. In total the network has 12 convolutional layers and 1.4 million

parameters.

3.2.3 Post-processing

The prediction from the segmentation stage contains 3D patches with continuous

voxel intensity values that representing how confident is the network in the voxel

belonging to an IVD. The final segmentation mask for each patch is obtained by binary

thresholding with a threshold of 0.5, which means voxels that are predicted more likely

to be IVD voxels than background voxels are included in the segmentation mask. Then

the mask patches are assembled back to a 3D volume of the lower spine, with the

same size of the IVD dataset, using the IVD center locations from the localization stage

and zero-padding.

17
 Chapter 4: Results

4.1 Wound Segmentation Results

We describe the evaluation metrics and compare the segmentation performance

of our method with several popular and state-of-the-art methods. Our deep learning

model is trained with data augmentation and preprocessing. Extensive experiments

were conducted to investigate the effectiveness of our network. FCN-VGG-16 is a

popular network architecture for wound image segmentation [19] [39]. Thus, we

trained this network on our dataset as the baseline model. For fairness of comparison,

we used the same training strategies and data augmentation strategies throughout the

experiments.

4.1.1 2D Evaluation Metrics

To evaluate the segmentation performance, Precision, Recall, and the Dice

coefficient are adopted as the evaluation metrics [40]:

Precision: Precision shows the accuracy of segmentation. More specifically, Precision

measures the percentage of correctly segmented pixels in the segmentation and is

computed by:
𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
Precision =
𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

Recall: Recall also shows the accuracy of segmentation. More specifically, it measures

the percentage of correctly segmented pixels in the ground truth and is computed by:
𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
Recall =
𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑛𝑒𝑔𝑡𝑖𝑣𝑒𝑠

18
Dice coefficient (Dice): Dice shows the similarity between the segmentation and the

ground truth. Dice is also called F1 score as a measurement balancing Precision and

Recall. More specifically, Dice is computed by the harmonic mean of Precision and

Recall:
2 × 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
Dice =
2 × 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑛𝑒𝑔𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

4.1.2 Experiment setup

The deep learning model in the presented work was implemented in python with

Keras [41] and TensorFlow [42] backend. To speed up the training, the models were

trained on a 64-bit Ubuntu PC with an 8-core 3.4GHz CPU and a single NVIDIA RTX

2080Ti GPU. For updating the parameters in the network, we employed the Adam

optimization algorithm [43], which has been popularized in the field of stochastic

optimization due to its fast convergence compared to other optimization functions.

Binary cross entropy was used as the loss function and we also monitored Precision,

Recall and the Dice score as the evaluation matrices. The initial learning rate was set

to 0.0001 and each minibatch contained only 2 images for balancing the training

accuracy and efficiency. The convolutional kernels of our network were initialized with

HE initialization [44] to speed up the training process and the training time of a single

epoch took about 77 seconds. We used early stopping to terminate the training so that

the best result was saved when there was no improvement for more than 100 epochs

in terms of Dice score. Eventually, our deep learning model was trained for around

1000 epochs before overfitting.

19
 To evaluate the performance of the proposed method, we compared the

segmentation results achieved by our methods with those by FCN-VGG-16 [19] [39]

and SegNet [18]. We also added 2D U-Net [45] to the comparison due to its

outstanding segmentation performance on biomedical images with a relatively small

training dataset. The segmentation results predicted by our model are demonstrated

in Figure 4 along with the illustration of our post-processing method. Quantitative

results evaluated with the different networks are presented in Table 1 where bold

numbers indicate the best results among all four models. To better demonstrate the

accuracy of the models, the numbers shown in the table are the highest possible

number reached among various trainings.

Table 1. Evaluation on our dataset.
Model VGG16 SegNet U-Net MobileNetV2 MobileNetV2+CCL

Precision 83.91% 83.66% 89.04% 90.86% 91.01%

Recall 78.35% 86.49% 91.29% 89.76% 89.97%

Dice 81.03% 85.05% 90.15% 90.30% 90.47%

4.1.3 Comparing our method to the others

In the performance measures, the highest Dice score was obtained by our

MobileNetV2+CCL model. VGG16 was shown to have the worst performance among

all the other CNN architectures. Our model also had the highest Precision of 94.76%,

which indicates that MobileNetV2 can segment the region with most true positive

pixels. The Recall of our model tested to be the second highest among all models, at

20
89.97%. This was 1.32% behind the highest Recall, 91.29%, which was achieved by

VGG16. Finally, the results show that our model achieves the highest overall accuracy

with a mean Dice score of 90.47%. Our accuracy was slightly better than the U-Net

architecture and significantly higher than SegNet and VGG16.

Comparing our model to VGG16, the Dice score is boosted from 81.03% to 90.47%

tested on our dataset. Based on the appearance of chronic wounds, we know that

wound segmentation is complicated by various shapes, colors and the presence of

different types of tissue. The patient images captured in clinic settings also suffer from

various lighting conditions and perspectives. In MobileNetV2, the deeper architecture

has more convolutional layers than VGG16, which makes MobileNetV2 more capable

to understand and solve these variables. MobileNetV2 utilizes residual blocks with skip

connections instead of the conventional convolution layers in VGG16 to build a deeper

network. These skip connections bridging the beginning and the end of a convolutional

block allows the network to access earlier activations that weren’t modified in the

convolutional block and enhance the capacity of the network.

Another comparison between U-Net and SegNet indicates that the former model

is significantly better in terms of mean Dice score. Similar to the previous comparison,

U-Net also introduces skip connections between convolutional layers to replace the

pooling indices operation in the architecture of SegNet. These skip connections

concatenate the output of the transposed convolution layers with the feature maps

from the encoder at the same level. Thus, the expansion section which consists of a

21
large number of feature channels allows the network to propagate localization

combined with contextual information from the contraction section to higher resolution

layers. Intuitively, in the expansion section or “decoder” of the U-Net architecture, the

segmentation results are reconstructed with the structural features that are learned in

the contraction section or the “decoder”. This allows the U-Net to make predictions at

more precise locations.

Besides the performance, our method is also efficient and lightweight. As shown

in Table 2, the total number of trainable parameters in MobileNetV2 was only a fraction

of the number in U-Net and VGG16. Thus, the network took less time during training

and could be applied to devices with less memory and limited computational power.

Alternatively, higher-resolution input images could be fed into MobileNetV2 with less

memory size and computational power comparing to the other models.

Table 2. Comparison of total numbers of trainable parameters.

Model VGG16 SegNet U-Net MobileNetV2

Number of
134,264,641 902,561 4,834,839 2,141,505
parameters

4.1.4 Comparison within the Medetec Dataset

Apart from our dataset, we also conducted experiments on the Medetec Wound

Dataset [46] and compared the segmentation performance of these methods. The

results are shown in Table 3. We directly applied our model, trained on our dataset, to

perform segmentation on the Medetec dataset as the testing dataset. The highest Dice

score is evaluated to 86.95% using MobileNetV2+CCL. This performance evaluation

22
agrees with our conclusion drawn from the testing on our dataset. MobileNetV2

outperforms the other models regardless of which chronic wound segmentation

dataset is used, thereby demonstrating that our model is robust and unbiased.
Table 3. Evaluation on the Medetec dataset.
Model VGG16 SegNet U-Net MobileNetV2 MobileNetV2+CCL

Precision 81.95% 74.39% 84.70% 84.25% 84.40%

Recall 85.41% 76.90% 86.40% 90.44% 90.65%

Dice 83.65% 75.62% 85.54% 87.23% 87.40%

4.2: IVD Segmentation Results

4.2.1 3D Evaluation Metrics

To evaluate the segmentation performance, two metrics are adopted from the

2015 MICCAI Challenge [25]. In addition to Dice coefficient mentioned in section 4.1.1,

we also calculated the Hausdorff distance (HD) that measures the distance between

two surface meshes. We compute HD for surfaces reconstructed from the ground true

segmentation mask and our segmentation result. Surfaces are generated using

Iso2mesh [47] from binary segmentation masks. The closest distance from each

vertex on the source surface mesh to the target surface mesh is found and HD is then

computed. A smaller HD value indicates better segmentation performance.

4.2.2 Effectiveness of the multi-modality data

The results show that the segmentation results achieved by excluding the in-phase

23
images from the training dataset are more accurate and less noisy near the lower IVDs

than that by the original full-modality data. Moreover, using the training dataset without

in-phase images, our localization network is able to learn more details and make much

more accurate predictions about the IVD centers. This makes the localization of

centers more stable and allows us to simply remove small regions (marked by yellow

boxes) and then crop a fix-size 3D patch for each IVD in the volume to train the

segmentation network.

From the multi-modality analysis, we found that the fat and in-phase images have

a significantly lower contrast among all the modalities. To analyze the effectiveness of

the multi-modality input data, we train our 3D network on 4 different combinations of

input modalities: 1) we train the network on full-modality images as the baseline, 2) we

exclude the fat images from all 4 modalities to build the second training dataset, 3) the

fat images are excluded from all 4 modalities, and 4) we only include oppose-phase

and water images in the last training dataset. The mean Dice scores of the

segmentation results predicted by the network trained on each dataset are presented

in Table 4. Among all the different training settings, the network trained on full-modality

images shows the worst segmentation performance. The reason is that the fat and in-

phase images have a lower contrast, which means that the input values of the network

are closer to each other and make it more difficult for the convolutional kernels to

distinguish between them. It is worthy of pointing out that input normalization does not

help with this situation because it is performed over the values of all the modalities. In

24
other words, if we treat these 4 types of images equally during the training process,

the fat and in-phase images confuse the network with their low image contrast.

Training dataset Combination Mean Dice ± SD

1) opp, wat, fat, and inn 87.9 ± 1.7

2) opp, wat, and fat 89.0 ± 1.4

3) opp, wat, and inn 88.0 ± 1.6

4) opp, and wat 88.5 ± 1.6

Table 4. Segmentation performance of our 3D method using different combinations of modalities as

the training dataset

4.2.3 Comparison of our methods and state-of-the-art methods

To evaluate the performance of the proposed methods, we compare the

segmentation results achieved by our methods with those by 3D U-Net[15], the CNN-

based team UNICHK [23] and the winning team UNIJLU [12] in the test1 dataset of

the 2015 MICCAI Challenge [25]. Quantitative results evaluated with the different

architectures are presented in Table 5. The mean Dice score obtained by our 3D

method is 89.0% with a standard deviation (SD) of 1.4%. We bring a 1.5% boost

comparing to the conventional 3D U-Net by training our network on 3D image patches

extracted from opposed-phase, water, and fat images. This result is still 2.5% behind

the state-of-the-art performance achieved by UNIJLU. The Mean HD of our 3D method

reached 0.8 mm with an SD of 0.3 mm, which indicates that our method is slightly

better when the segmentation results are reconstructed to 3D models. The strength of

deep learning methods is the computation time. The Theano-based implementation of

25
3D U-Net from UNICHK takes 3.1s to process one 40 × 512 × 512 volume. Our

network is implemented based on TensorFlow and it takes about 0.5s to segment all

the IVDs in a 36 × 256 × 256 input volume. Overall, the computation time of our end-

to-end segmentation is about 10s including localization, preprocessing, segmentation

and postprocessing. Whereas it takes 5 min on average to segment all IVDs for a

patient by UNIJLU. It is also worth mentioning that the training dataset used in our

study only contains data from 6 patients while UNICHK and UNIJLU have access to a

training dataset from 16 patients i.e. our network is able to learn the 3D geometric

morphometrics of IVDs with much fewer data to learn from.

Table 5. Segmentation result evaluation of the conventional 3D U-Net, UNICHK, UNIJLU and our
method.
Methods Mean Dice ± SD Mean HD ± SD

3D U-Net 87.5 ± 0.9 1.1 ± 0.2

UNICHK 88.4 ± 3.7 1.3 ± 0.2
UNIJLU 91.5 ± 2.3 1.1 ± 0.2
Our method 89.0 ± 1.4 0.8 ± 0.3

Chapter 5: Conclusions

We attempted to solve two problems using deep learning: 1) the automated

segmentation of chronic foot ulcers in a dataset we built on our own. 2) The automated

segmentation of IVDs from 3D MRI scans. For evaluating the performance, we

conducted comprehensive experiments and analyses on SegNet, VGG16, 2D U-Net,

3D U-Net and our model based on modified 3D U-Net and another proposed model

based on MobileNetV2 and CCL. In the comparison of various neural networks, our

26
methods have demonstrated their effectiveness and in the field of medical image

segmentation due to their fully convolutional architectures. We also demonstrated the

robustness of our models by testing it on publicly available datasets where our model

still achieves the highest Dice score. In the future, we plan to improve our work by

extracting the shape features separately from the pixel-wise convolution in the deep

learning model. Also, we will include more data in the dataset to improve the

robustness and prediction accuracy of our method.

References

[1] Song, Bo, and Ahmet Sacan. Automated wound identification system based on

image segmentation and artificial neural networks. In 2012 IEEE International

Conference on Bioinformatics and Biomedicine. 1-4 (2012).

[2] A. I. Lopez, B. Glocker, “Complementary classification forests with graph-cut

refinement for ivd localization and segmentation,” in Proc. the 3rd MICCAI

Workshop & Challenge on Computational Methods and Clinical Applications for

Spine Imaging, 2015.

[3] Fauzi, Mohammad Faizal Ahmad et al. Computerized segmentation and

measurement of chronic wound images. Computers in biology and medicine. 60,

74-85 (2015).

[4] D. Forsberg, "Atlas-based registration for accurate segmentation of thoracic and

lumbar vertebrae in CT data," in Recent Advances in Computational Methods

and Clinical Applications for Spine Imaging, pp. 49-59, Springer, Cham, 2015.

27
[5] C. Chen, D. Belavy, W. Yu, C. Chu, G. Armbrecht, M. Bansmann, D. Felsenberg,

and G. Zheng, "Localization and segmentation of 3D intervertebral discs in MR

images by data driven estimation," IEEE transactions on medical imaging, vol.

34, no. 8, pp. 1719-1729, 2015.

[6] Hettiarachchi, N. D. J., R. B. H. Mahindaratne, G. D. C. Mendis, H. T.

Nanayakkara, and Nuwan D. Nanayakkara. Mobile based wound measurement.

In 2013 IEEE Point-of-Care Healthcare Technologies (PHT). 298-301 (2013).

[7] Z. Wang, X. Zhen, K.Y. Tay, S. Osman, W. Romano, and S. Li. "Regression

segmentation for M3 spinal images," IEEE transactions on medical imaging, vol.

34, no. 8, 1640-1648, 2015.

[8] Hani, Ahmad Fadzil M., Leena Arshad, Aamir Saeed Malik, Adawiyah Jamil, and

Felix Yap Boon Bin. Haemoglobin distribution in ulcers for healing assessment.

In 2012 4th International Conference on Intelligent and Advanced Systems

(ICIAS2012). 1, 362-367 (2012).

[9] Wantanajittikul, Kittichai, Sansanee Auephanwiriyakul, Nipon Theera-Umpon,

and Taweethong Koanantakool. Automatic segmentation and degree

identification in burn colour images. In The 4th 2011 Biomedical Engineering

International Conference. 169-173 (2012).

[10] H. Hutt, R. Everson, and J. Meakin, "3d intervertebral disc segmentation from

MRI using supervoxel-based crfs," in International Workshop on Computational

Methods and Clinical Applications for Spine Imaging, pp. 125-129, Springer,

28
 Cham, 2015.

[11] M. Urschler, K. Hammernik, T. Ebner, and D. Štern, "Automatic intervertebral disc

localization and segmentation in 3d mr images based on regression forests and

active contours," In International Workshop on Computational Methods and

Clinical Applications for Spine Imaging, pp. 130-140, Springer, Cham, 2015.

[12] R. Korez, B. Ibragimov, B. Likar, F. Pernuš, and T. Vrtovec. "Deformable model-

based segmentation of intervertebral discs from MR spine images by using the

SSC descriptor," in International Workshop on Computational Methods and

Clinical Applications for Spine Imaging, pp. 117-124, Springer, Cham, 2015.

[13] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification

with deep convolutional neural networks. In Advances in neural information

processing systems. 1097-1105 (2012).

[14] Russakovsky, Olga et al. Imagenet large scale visual recognition challenge.

International journal of computer vision. 115, no. 3, 211-252 (2015).

[15] Garcia-Garcia, Alberto, Sergio Orts-Escolano, Sergiu Oprea, Victor Villena-

Martinez, and Jose Garcia-Rodriguez. A review on deep learning techniques

applied to semantic segmentation. arXiv preprint arXiv:1704.06857 (2017).

[16] LeCun, Yann, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based

learning applied to document recognition. Proceedings of the IEEE. 86, no. 11,

2278-2324 (1998).

[17] Long, Jonathan, Evan Shelhamer, and Trevor Darrell. Fully convolutional

29
 networks for semantic segmentation. In Proceedings of the IEEE conference on

computer vision and pattern recognition. 3431-3440 (2015).

[18] Wang, Changhan et al. A unified framework for automatic wound segmentation

and analysis with deep convolutional neural networks. In 2015 37th annual

international conference of the IEEE engineering in medicine and biology society

(EMBC). 2415-2418 (2015).

[19] Goyal, Manu, Moi Hoon Yap, Neil D. Reeves, Satyan Rajbhandari, and Jennifer

Spragg. Fully convolutional networks for diabetic foot ulcer segmentation. In

2017 IEEE international conference on systems, man, and cybernetics (SMC).

618-623 (2017).

[20] Liu, Xiaohui et al. A framework of wound segmentation based on deep

convolutional networks. In 2017 10th International Congress on Image and

Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI). 1-7

(2017).

[21] H. Chen, Q. Dou, X. Wang, J. Qin, J. CY Cheng, and P. A. Heng, "3D fully

convolutional networks for intervertebral disc localization and segmentation," in

International Conference on Medical Imaging and Virtual Reality, pp. 375-382,

Springer, Cham, 2016.

[22] X. Li, Q. Dou, H. Chen, C. Fu, X. Qi, D. L. Belavý, G. Armbrecht, D. Felsenberg,

G. Zheng, and P. A. Heng, "3D multi-scale FCN with random modality voxel

dropout learning for Intervertebral Disc Localization and Segmentation from

30
 Multi-modality MR Images," Medical image analysis, vol. 45, pp. 41-54, 2018.

[23] H. Chen, Q. Dou, X. Wang, P. A. Heng, “Deepseg: Deep segmentation network

for intervertebral disc localization and segmentation,” in Proc. 3rd MICCAI

Workshop & Challenge on Computational Methods and Clinical Applications for

Spine Imaging, 2015.

[24] Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ronneberger, "3D U-

Net: learning dense volumetric segmentation from sparse annotation," in

International Conference on Medical Image Computing and Computer-Assisted

Intervention, pp. 424-432. Springer, Cham, 2016.

[25] G. Zheng, C. Chu, D. L. Belavý, B. Ibragimov, R. Korez, T. Vrtovec, and H. Hutt

et al, "Evaluation and comparison of 3D intervertebral disc localization and

segmentation methods for 3D T2 MR data: A grand challenge," Medical image

analysis, vol. 35, pp. 327-344, 2017.

[26] S. Kim, W. Bae, K. Masuda, C. Chung, and D. Hwang, "Fine-grain segmentation

of the intervertebral discs from MR spine images using deep convolutional neural

networks: BSU-Net," Applied Sciences, vol. 8, no. 9, pp. 1656, 2018.

[27] J. Dolz, C. Desrosiers, and I. B. Ayed, "IVD-Net: Intervertebral disc localization

and segmentation in MRI with a multi-modal Unet," arXiv preprint

arXiv:1811.08305, 2018.

[28] J. T. Lu, S. Pedemonte, B. Bizzo, S. Doyle, K. P. Andriole, M. H. Michalski, R. G.

Gonzalez, and S. R. Pomerantz. "DeepSPINE: Automated Lumbar Vertebral

31
 Segmentation, Disc-level Designation, and Spinal Stenosis Grading Using Deep

Learning," arXiv preprint arXiv:1807.10215, 2018.

[29] Sandler, Mark, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-

Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In

Proceedings of the IEEE conference on computer vision and pattern recognition.

4510-4520 (2018).

[30] C. Chen, D. Belavy, and G. Zheng, "3D intervertebral disc localization and

segmentation from MR images by data-driven regression and classification," in

International Workshop on Machine Learning in Medical Imaging, pp. 50-58,

2014.

[31] Redmon, Joseph, and Ali Farhadi. Yolov3: An incremental improvement. Preprint

at arXiv:1804.02767 (2018).

[32] Tzutalin. LabelImg. Git code https://github.com/tzutalin/labelImg (2015).

[33] Chollet, François. Xception: Deep learning with depthwise separable

convolutions. In Proceedings of the IEEE conference on computer vision and

pattern recognition. 1251-1258 (2017).

[34] Nair, Vinod, and Geoffrey E. Hinton. Rectified linear units improve restricted

boltzmann machines. In Proceedings of the 27th international conference on

machine learning (ICML-10). 807-814 (2010).

[35] Krizhevsky, Alex, and Geoff Hinton. Convolutional deep belief networks on cifar-

10. Unpublished manuscript. 40, 1-9 (2010).

32
[36] Everingham et al. The pascal visual object classes challenge: A retrospective.

International journal of computer vision. 111, no. 1, 98-136 (2015).

[37] Pearce, David J. An improved algorithm for finding the strongly connected

components of a directed graph. Victoria University, Wellington, NZ, Tech. Rep

(2005).

[38] S. Park, and N. Kwak, "Analysis on the dropout effect in convolutional neural

networks," in Asian Conference on Computer Vision, pp. 189-204. Springer,

Cham, 2016.

[39] Li, Fangzhao, Changjian Wang, Xiaohui Liu, Yuxing Peng, and Shiyao Jin. A

composite model of wound segmentation based on traditional methods and deep

neural networks. Computational intelligence and neuroscience. (2018).

[40] Zou, Kelly H. et al, Statistical validation of image segmentation quality based on

a spatial overlap index1: scientific reports. Academic radiology. 11, no. 2, 178-

189 (2004).

[41] F. Chollet et al, Keras, chollet2015keras, https://keras.io.

[42] S. S. Girija, Tensorflow: Large-scale machine learning on heterogeneous

distributed systems. (2016).

[43] Kingma, Diederik P., and Jimmy Ba. Adam: A method for stochastic optimization.

Preprint at arXiv:1412.6980 (2014).

[44] He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into

rectifiers: Surpassing human-level performance on imagenet classification. In

33
 Proceedings of the IEEE international conference on computer vision. 1026-1034

(2015).

[45] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. U-net: Convolutional

networks for biomedical image segmentation. In International Conference on

Medical image computing and computer-assisted intervention. 234-241 (2015).

[46] Thomas, Stephen. Stock Pictures of Wounds. Medetec Wound Database

http://www.medetec.co.uk/files/medetec-image-databases.html (2020).

[47] Q. Fang, and D. A. Boas. "Tetrahedral mesh generation from volumetric binary

and grayscale images," in Biomedical Imaging: From Nano to Macro, ISBI'09,

IEEE International Symposium, pp. 1142-1145, 2009.

34