• No results found

Tracking and Compensation for Breast Deformations during Image-Guided Breast Biopsy with a Hand-Mounted Biopsy Tool

N/A
N/A
Protected

Academic year: 2021

Share "Tracking and Compensation for Breast Deformations during Image-Guided Breast Biopsy with a Hand-Mounted Biopsy Tool"

Copied!
62
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

TRACKING AND COMPENSATION FOR BREAST DEFORMATIONS DURING IMAGE-GUIDED BREAST BIOPSY WITH A HAND-MOUNTED BIOPSY TOOL

J. (Jos) Tichelaar

MSC ASSIGNMENT

Committee:

prof. dr. ir. S. Stramigioli dr. V. Groenhuis, MSc dr. F.J. Siepel, MSc dr. ir. L.J. Spreeuwers

Maart

, 2021 012RaM2021 Robotics and Mechatronics

EEMCS University of Twente P.O. Box 217 7500 AE Enschede

(2)

Abstract

In breast screening lesions might be discovered which have to be checked for ma- lignancy. To do so, a biopsy can be performed in which a needle is inserted into the breast to recover sample cells from the lesion for further research. This biopsy is usually performed under the guidance of an imaging modality. For MRI-guided biopsies, the patient has to move in between getting the scan and having the biopsy, due to the limited space inside the bore and the limitations induced by the strong magnetic field. Two options are discussed: a robotic system inside the MRI scanner and a robotic system outside the MRI scanner, the latter of which will be used in this assignment. Between making the MRI scan and getting the biopsy, the patient has to move, which might mean that the lesion moves due to for instance breathing or muscle relaxation. Real-time imaging can be used to update the location of the lesion to help the doctor target the lesion more accurately.

The real-time imaging is done by putting markers on the breast, which are tracked using an OptiTrack system. The location of the markers is used in combination with the marker locations in the MRI scan to determine the current breast configuration and lesion location using the Thin Plate Spline algorithm. Algorithms have been im- plemented to label the markers to their previous frames and thus also to the marker locations in the MRI scan. Simulation results have been used to find an optimal combination of algorithms, which is explained to be the distance, angle, and height algorithms. The algorithms have been implemented and experiments are performed with different marker sizes and number of markers to find their influence on the total performance of the system. Using 8 mm markers no biopsies could be performed, which was achievable with marker sizes of 12, 16, and 20 mm. Biopsies could be performed for marker totals up to and including 9 (and 8 for 16 mm markers), where jumping markers were observed for the greatest part when using more markers. From the results, it is concluded that the size of the marker is of lesser importance than the number of markers. A follow-up experiment is suggested in which the validation of the lesion location is examined using the different marker sizes and the number of markers.

All performed biopsies have a sub-millimeter accuracy, which is better than the average accuracy that Marta reached in her research (2.21 mm). When comparing these results to a studies performed in 2009, in which 20 biopsies were performed on breast phantom and 32 biopsies on patients’ breasts, a clear increase in biopsy accuracy can be seen.

(3)

Acknowledgement

First of all, I would like to thank my day-to-day supervisor dr. Vincent Groenhuis.

The meetings every Monday in combination with fast email responses helped me with every problem I encountered and headed me in the right direction. Besides, I want to thank everyone that has ever been present at the general meetings online during my master’s assignment. Hearing potential problems and suggestions from people with a completely different perspective helped me to substantiate every claim and understand the basics better. Finally, I would like to thank my parents for keeping me motivated and assisting me with the makings of the markers.

(4)

Contents

List of Figures v

List of Tables vii

1 Introduction 1

1.1 Screening methods . . . . 1

1.1.1 Mammography . . . . 1

1.1.2 Ultrasonography . . . . 2

1.1.3 Magnetic Resonance Imaging . . . . 3

1.1.4 Biopsy . . . . 4

1.2 Robotic solutions . . . . 5

1.2.1 Robotic system inside the MRI scanner . . . . 6

1.2.2 Robotic system outside the MRI scanner . . . . 6

1.2.2.1 Motion capture possibilities . . . . 6

1.2.2.2 Markers . . . . 6

1.3 Main research question . . . . 7

1.3.1 Approach . . . . 7

2 Method 8 2.1 Markers . . . . 8

2.1.1 MR images . . . . 8

2.1.2 Design . . . . 8

2.1.2.1 Attachment system . . . . 9

2.1.2.2 Markers . . . . 10

2.1.2.2.1 Outside of the marker . . . . 10

2.1.2.2.2 Inside of the marker . . . . 10

2.1.3 Petroleum jelly . . . . 11

2.1.4 Quantity . . . . 11

2.1.5 Retro-reflective tape . . . . 12

2.2 MRI processing . . . . 13

2.2.1 MRI scan . . . . 13

2.2.2 Analysis . . . . 13

2.2.2.1 Coordinate system . . . . 13

2.2.2.2 Markers and 3D model of the breast . . . . 14

2.2.2.3 Lesions . . . . 15

2.3 Needle-steering guide . . . . 17

2.3.1 Base . . . . 17

2.3.2 Needle-holder . . . . 18

(5)

CONTENTS CONTENTS

2.3.2.1 Biopsy needle . . . . 19

2.3.2.2 Obtaining needle-tip position . . . . 21

2.3.2.3 Validation of needle-tip position . . . . 23

2.4 Lesion tracking . . . . 24

2.4.1 Validation results . . . . 24

2.4.2 Theory of TPS . . . . 24

2.4.3 Deformation of the breast . . . . 26

2.5 OptiTrack . . . . 29

2.5.1 Software . . . . 29

2.5.2 Marker overlap . . . . 31

3 Marker identification 33 3.1 Previous research . . . . 34

3.2 Implementation . . . . 35

3.2.1 Calibration . . . . 40

3.2.2 Inter-frame identification . . . . 41

3.2.3 Marker occlusion . . . . 42

3.2.4 Simulation results using real data . . . . 42

4 Experiments 44 4.1 Setup . . . . 44

4.2 Results . . . . 47

4.3 Discussion . . . . 49

5 Conclusion 50 5.1 Future recommendations . . . . 51

Bibliography 52

(6)

List of Figures

1.1 Detection of breast cancer using mammography [1]. Two lumps have

been identified and encircled in yellow. . . . 2

1.2 Ultrasound image of a breast [2] . . . . 3

1.3 MR image of a breast [3] . . . . 4

1.4 Ultrasound-guided biopsy [4] . . . . 5

2.1 Top, front, and isometric view of part one of the attachment system . 9 2.2 Top, front, and isometric view of part two of the attachment system . 9 2.3 Part one on the left, part two in the middle, and the connection of the two parts on the right . . . . 10

2.4 SolidWorks outer-marker designs for marker sizes 20, 16, 12 and 8 mm (from left to right) . . . . 10

2.5 Markers from figure 2.4 under a different angle to show the inner markers 11 2.6 Side view of the phantom with all 14 markers on it . . . . 12

2.7 3D model of the phantom in the prone position showing the location of the center of the phantom and some of the markers. . . . 14

2.8 Screen capture of a phantom slice in the three planes and 3D view on top right . . . . 15

2.9 Needle-steering guide [5] . . . . 17

2.10 SolidWorks design of the first part of the needle holder (schematic) [5] 18 2.11 Three different views of the second part of the needle-holder . . . . 19

2.12 Four different phases of the biopsy needle mechanism . . . . 20

2.13 Coordinate systems used on the Aimbot device . . . . 22

2.14 Phantom in the original configuration. The visible phantom markers are in blue, while the markers that are behind the phantom are in transparent blue. The location of the lesion is red. A force vector F is drawn to indicate the direction of the applied force to achieve the phantom shape in figure 2.15. . . . 27

2.15 Phantom shape after applying the force as indicated in figure 2.14 . . 27

2.16 2D image acquired by one of the cameras. 6 markers can be identified on the image. . . . 30

(7)

LIST OF FIGURES LIST OF FIGURES

2.17 Example image captured by one of the cameras showing marker over- lap. Around the center of the image two markers can be seen which are overlapping. They get classified as one marker with a rather strange shape, which does not meet the roundness requirements. The red cross in combination with the label ’Circle Filter’ indicates that this combi- nation of markers will not be classified as a marker and can thus also not be used in the calculation of the marker centroids for the merged markers. . . . 32 3.1 Example of an inter-frame marker identification problem. Without any

additional information it is currently unclear whether the red marker is at the top or bottom in frame 2. . . . 33 4.1 Screenshot of the GUI options, showing the options to activate both

models and the azimuth, elevation, and distance values that belong to the scene from figure 4.2 . . . . 45 4.2 Top view of the scene, in which it can be seen that the needle is inserted

in the phantom . . . . 45 4.3 Right side view of the scene in which the phantom model and Aimbot

model (without the base) are shown . . . . 46 4.4 Setup of the lab showing the plateaus, all cameras, and the OptiTrack

computer. The red capsules indicate the location of the cameras with their corresponding numbers. Two numbers at a capsule indicate that at that location there are two cameras with a different orientation. . . 46 4.5 Screenshot of one of the cameras’ 2D image showing ten 8 mm markers 47

(8)

List of Tables

2.1 Settings used to create the MRI scan . . . . 13 2.2 LPS coordinates of the 14 markers on the phantom during imaging . . 15 2.3 Coordinates of the lesions in RAS and LPS frames . . . . 16 2.4 Needle characteristics . . . . 19 2.5 Needle tip validation results on 6 markers . . . . 23 2.6 List of locations of the markers and the lesion before and after applying

the indicated force of figure 2.14. The last column indicates how much the certain object has moved in between the two configurations . . . . 28 3.1 Data from two consecutive frames. The left marker set (frame i) has

already been labeled using the frame before that (frame i − 1), where the first 6 markers are the phantom markers and the last 4 markers are Aimbot markers. The right marker set (frame i + 1) will be labeled using the left data set. . . . 35 3.2 Simulation results of the different combinations of algorithms on 1

dataset. The crosses in the first 4 columns indicate which of the algo- rithms (or combination of them) were used. The 5th column indicates the total amount of frames for which the algorithms were able to cal- culate a final output using 2 runs of the output array (example in equations 3.14 and 3.15). The 6th column indicates the number of frames from the frames with a distinct output for which the labeling was incorrect. For the last column the frequency of the algorithms has been calculated (averaged over 5) . . . . 43 4.1 Set of selected markers from table 2.2 . . . . 47 4.2 Experiment results with 12 mm markers. Biopsies are performed up to

and including 9 markers . . . . 48 4.3 Experiment results with 16 mm markers. Biopsies are performed up to

and including 8 markers . . . . 48 4.4 Experiment results with 20 mm markers. Biopsies are performed up to

and including 9 markers . . . . 48

(9)

Chapter 1

Introduction

Breast cancer is the most commonly diagnosed form of cancer in women worldwide.

In 2012 twenty-five percent of all cancer cases were identified as breast cancer [6].

Breast cancer might develop due to some of the breast cells growing abnormally.

The cancerous cells divide faster than the healthy cells, increasing in size and thus forming lumps. Early detection of breast cancer is one of the major reasons for the increase in survival rate [7].

1.1 Screening methods

In the Netherlands, the National Breast Cancer Screening Programme has been set up. This program is created for women between 50 and 75 inviting them for a mam- mogram every two years. In 2010 approximately a million women participated in breast screening, wherein 6600 cases of breast cancer were detected [8]. The most common screening methods are mammography, ultrasonography and magnetic reso- nance imaging [9].

1.1.1 Mammography

During the mammography procedure, the breast of the patient is placed on a platform so it can be compressed using a plastic paddle. This compression is required in order to flatten out the breast for better visualization. The tissue is spread out decreasing the chance of abnormalities, such as lesions, to be hidden by overlying tissue. The amount of breast tissue that is being imaged is decreased which means that a lower amount of x-ray dose can be used. Movement of the breast will be minimized, as the breast is fixed between the platform and the plastic paddle. Less movement of the breast means less blurring of the images that are returned.

The patient is asked not to breathe when the x-ray pictures are taken in order to not blur the images. When the x-ray machine is positioned correctly a small burst of radiation is produced which passes through the breast. The x-rays are absorbed in different levels depending on the density of a certain part of the breast. Dense bones absorb much of the radiation compared to soft tissue such as for instance muscles where most of the x-rays pass through. Consequently, in the mammogram bones show up in white, soft tissue in grey, and air in black, from which an example can be

(10)

seen in figure 1.1. Two mammograms are displayed both showing a lump encircled in yellow, which indicates the presence of cancerous cells.

Figure 1.1: Detection of breast cancer using mammography [1]. Two lumps have been identified and encircled in yellow.

The images are analyzed by a radiologist and the evaluation is sent to the patient’s physician who will discuss the results with the patient. Unfortunately, there are some limitations to mammography. One major drawback is that some mammograms are wrongly identified. Approximately half of the women who get annual mammograms deal with a false-positive mammogram [10]. False negatives also happen as not all breast cancers can be seen using mammography [11].

Conventional mammography uses 2D images, whereas digital tomosynthesis uses x-rays to create a 3D picture of the breast. The results of [12] indicate that digi- tal breast tomosynthesis (DBT) performs slightly better than the conventional two- dimensional digital mammography (DM). Both the false-negative rate and the false- positive rate have decreased, whereas the cancer detection rate has increased.

1.1.2 Ultrasonography

Ultrasonography (ultrasounds) uses high-frequency sound waves to create an image of the breast. The doctor moves a probe over the breast, which sends those waves into the breast. The waves bounce back from the breast tissue and are received by the probe. The received waves are used to create the image.

Ultrasound is often used as a complimentary examination if an abnormality has been found during mammography. This is done to check if the abnormality can be positively identified as a tumor or something else. On its own, ultrasound is not used in breast screening as it can miss early signs of cancer. There are some exceptions for using ultrasound over mammography including very dense breasts and pregnancy.

The images that are obtained from ultrasound are examined, similar to mammography [13]. Figure 1.2 shows an example image of a breast scanned using ultrasound.

(11)

Figure 1.2: Ultrasound image of a breast [2]

1.1.3 Magnetic Resonance Imaging

If the aforementioned screening methods, mammography and ultrasound, do not clearly show present lesions in the breast it might be opted to use Magnetic Res- onance Imaging (MRI). This screening method makes use of the presence of hydrogen protons present in water and fat in the body. Naturally, those protons spin in the body with a random axis alignment. When the patient is moved inside the bore a strong magnetic field works on the body which aligns those spin axes, creating a mag- netic vector along the axis of the MRI scanner. Thereafter a radio wave can be added to the magnetic field causing the hydrogen nuclei to resonate. If the radio frequency source is switched off the magnetic vector returns back to its initial state, emitting a radio wave signal. This signal is received by receiver coils and its intensity is used to produce MR images, such as the one in figure 1.3 [14].

The strong magnetic field induced by the MR machine requires some additional care. Any present ferromagnetic objects are attracted by the field and might move with great force because of it. External objects such as bracelets, cell phones and so on can simply be removed from the room. This, however, is more difficult for objects that reside in the body, such as metallic implants or heart pacemakers. Protocols are put in place to check if the procedure can still be performed [15]. The Food and Drug Administration (FDA) defines three MRI safety labels: MR unsafe, MR conditional, and MR safe. Items that are labeled MR safe are of no risk for the MRI procedure and do not have to be moved outside of the room. MR conditional items may enter the room if the conditions that have been set for them are realized. The purpose of the device at issue determines the conditions for safe use. MR unsafe items should in no case enter the operation room [16].

(12)

Figure 1.3: MR image of a breast [3]

1.1.4 Biopsy

A biopsy will be performed if during screening a lesion has been discovered in the breast. A needle is inserted and a sample of the abnormal area is removed for lab- oratory testing [4]. If the biopsy is performed and the lesion is missed then healthy sample cells are retrieved from the breast and the biopsy has to be redone. To reduce this false biopsy rate it is important to accurately know the lesion location. The biopsy is therefore often performed under ultrasound or MRI guidance. Figure 1.4 demonstrates the procedure of an ultrasound-guided biopsy. A mass (red) has been detected during screening and a sample of it will be removed using the biopsy needle.

The ultrasound probe is positioned such that a clear image of the mass and the needle tip is obtained. When the mass has been reached the biopsy gun fires and extracts the tissue.

For MRI-guided biopsies a bed with a hole for the breast, which will be clamped by two vertical plates (of which one has a rectangular grid on it), is used. The patient’s breast is scanned using the MRI machine, after which the required grid position and needle insertion depth are calculated. The patient moves out of the scanner and a stylet is inserted to get access to the lesion. The stylet is replaced by an obturator and the patient is scanned again to check if the tip location corresponds with the location of the lesion. If the two locations correspond, then is the patient moved out of the scanner to insert the biopsy needle.

A major drawback to the current MRI-guided biopsy procedure is the lack of real- time feedback. The patient is moved inside the MRI machine to create a scan of the breast and a projected path is calculated. The biopsy, however, is performed out- side the MRI machine, due to the confined space inside the MRI scanner, increasing the difficulty of performing a biopsy under visual guidance. In between scanning the breast and inserting the needle according to the projected path, the lesion might have moved. Varying load conditions, such as breathing, muscle contraction, or needle- tissue interaction, could ensure that the lesion moves slightly, even if the breast is immobilized by the two vertical plates [17].

(13)

Figure 1.4: Ultrasound-guided biopsy [4]

The current MRI-guided biopsy procedure could benefit from a real-time feedback system that is capable of updating the location of the lesions. In [18] a study has been performed to determine the spatial localization errors in MRI-guided biopsies on both phantoms and patients. Three different trained system operators in total performed 20 biopsies on 6 different phantoms and 32 biopsies on 22 patients. The localization error, expressed as the euclidean distance between needle tip and lesion center, was 4.4 ± 2.9 mm for the phantoms and 5.7 ± 3.0 mm for the patients’ breasts.

1.2 Robotic solutions

A possible solution to the aforementioned problems in MRI-guided biopsies is the implementation of robotic systems. Two types of robotic systems for MRI-guided

(14)

breast biopsies can be characterized: a robotic system inside the MRI scanner and a robotic system outside the MRI scanner.

1.2.1 Robotic system inside the MRI scanner

The major advantage of a working robotic system inside the MRI scanner would be that the patient does not have to move in between imaging and the operation. The lesion can be targeted under (near) real-time guidance. A major drawback, however, is the fact that all components have to be MRI compatible, as has already been elaborated on in section 1.1.3. Another big disadvantage is that there is little room for such a robotic system inside the MRI scanner due to limited space. Ming Li et al.

have worked on an MRI-compatible robotic system for an aortic valve replacement inside the MRI scanner [19]. In his dissertation, Vincent Groenhuis describes five iterations of the Stormram research line, designed for performing biopsies inside the MRI scanner [17].

1.2.2 Robotic system outside the MRI scanner

For using a robotic system outside the MRI scanner, multiple steps have to be taken.

The first step is to put markers on the breast which will be used to track deformations of the breast and interpolate the locations of the lesions during the biopsy procedure.

Subsequently, an MRI scan is made from which a 3D patient-specific model of the breast is made. The MRI scan is also used to find the locations of the lesions and the markers. During the biopsy procedure, the markers will be tracked, with which the locations of the lesions can be updated based on the current marker configuration.

1.2.2.1 Motion capture possibilities

In the master thesis of Marta Lagomarsino four different tracking technologies are described: Mechanical, Acoustic, Optical, and Electromagnetic. Each of these tech- nologies is evaluated on four different types of performance: Accuracy, Real-time update rate, Robustness to interference, and motion constraint. Based on the result of these performances the conclusion is drawn for use of an optical tracking technique [5]. This master’s assignment will be a continuation of the work that she has per- formed at the University of Twente, and therefore also an optical tracking technique will be used.

Three different acquisition devices were considered in her thesis: a Kayeton stereo camera, a StereoPi (stereoscopic camera based on Raspberry Pi), and an OptiTrack system. Results indicate that the StereoPi is more reliable and robust than the Kayeton stereo camera, but also that the StereoPi unexpectedly works just as well or better than the OptiTrack system. This unexpected behavior can be substantiated by the erroneous identifications of markers 2.42 times per minute and the lengthier execution time.

It appears that the current system in combination with OptiTrack can be improved upon and will therefore be used in my research.

1.2.2.2 Markers

Two main types of markers can be identified when using OptiTrack: passive and ac- tive markers. The main difference between the two is that active markers emit light

(15)

whereas passive markers only reflect light [20]. Active markers thus require some kind of electrical system capable of emitting light using LEDs, which has to meet the MRI conditions. To make these markers yourself might be very difficult, especially since the markers have to be modified for them to be seen on the MRI scan. The downsides of active markers for this application leads to the use of passive markers.

To improve the current system some marker parameters will have to be investi- gated: marker size and the number of markers on the breast. Markus Windolf et al. researched the systematic accuracy and precision of the Vicon-460 system. Their results indicate that bigger markers produce higher accuracy and precision which was accomplished with markers of sizes 9.5, 15, and 25 mm [21]. Yongqiang Feng et al.

did research to find the accuracy and precision of an optical motion system in 2D and 3D during speech and non-speech motor tasks. Among other things, they performed experiments for static and dynamic marker tracking using markers of sizes 3 and 6 mm and varying between 60, 120, and 240 fps. For static marker tracking in 2D, bet- ter accuracy is obtained in both dimensions for all frame rates for the bigger marker.

The same line of results can be seen for static marker tracking in 3D, with some ex- ceptions (x-dimension for 120 and 240 fps, y-dimension for 60 fps and z-dimension for 120 fps). For dynamic marker tracking, similar results can be seen for 2D. However, for 3D dynamic marker tracking the result for the 3 mm marker are extremely close to the result for the 6 mm marker and even better for 60 fps [22].

The number of markers that can be used during tracking depends on multiple factors: the size of the breast, the size of the markers, and the spacing of the markers.

To ensure that the markers will be identified correctly a certain distance will be kept between markers. The number of markers will thus be chosen according to the desired inter-marker distance.

1.3 Main research question

The main research question is phrased as

How do marker size and the number of markers influence the accuracy of a biopsy performed on a deformable phantom using a 2DOF needle-steering guide by means of optical marker tracking?

The goal of this assignment is to combine pre-operative MR images with optical motion capture to interpolate the current lesion location and improve the biopsy accu- racy, measured in absolute distance between biopsy needle tip and lesion center. This assignment will be performed as part of the MRI and Ultrasound Robotic Assisted Biopsy (MURAB) program. In this assignment, a breast phantom will be used.

1.3.1 Approach

This dissertation will be divided into three main parts: In the first part, the required steps before marker tracking will be discussed. This is done in Chapter 2, where all necessities (the markers, MRI processing, the needle-steering guide, lesion tracking, and the OptiTrack system) will be elaborated. The second part of the dissertation will be about the implementation of the marker identification algorithms (Chapter 3) and the results that are obtained with it (Chapter 4). The final part will be the conclusion (Chapter 5).

(16)

Chapter 2

Method

This chapter will go into detail concerning the required knowledge for the experiment to be understood. Five main sections will be defined: Markers, MRI processing, Needle-steering guide, Lesion tracking, and OptiTrack.

2.1 Markers

In this section, all essentials about the markers will be discussed. The markers will be designed in SolidWorks and 3D printed using a Creality Ender 3 Pro.

2.1.1 MR images

The material that is used to print the markers is Polylactic Acid (PLA) as it is con- sidered MR safe concerning the static magnetic field of the MRI scanner. Besides, it can also be considered quite safe concerning gradient or RF-heating as its volume resistivity is determined to be two orders of magnitude larger than other known MR safe materials [23]. The material itself, however, is not visible on MR images, thus requiring the addition of something that both meets the MR compatibility require- ments and is visible on MR images. Petroleum jelly meets both these requirements and can easily be acquired and will thus be used to make the markers visible on MR images. In practice, a 5mm diameter bubble of petroleum jelly should yield recogniz- able objects in the MR images. The software that will be used to locate the center of the petroleum jelly bubbles can automatically detect them if there is a minimum gap distance of at least 2 mm between the breast and the petroleum jelly.

2.1.2 Design

Besides the markers, also some kind of attachment system has to be considered in order to be able to attach and remove markers from the phantom. The designed attachment system will consist of two parts, one of which will be glued to the phantom, and the other one on which the markers will be glued.

(17)

2.1.2.1 Attachment system

The first part of the attachment system is a disk with a diameter of 20 mm and a height of 4 mm. A squared extruded cut of 10x10 mm is made centered around the center of the disk, which goes all the way through. Figure 2.1 shows the top, front and isometric view of part one of the attachment system. The part of the disk that remains after the extruded cut in the top view will be glued on the phantom.

Figure 2.1: Top, front, and isometric view of part one of the attachment system The second part of the attachment system resembles the first part of the attachment but now has a square-shaped extruded base. This extruded base is supposed to slide into the extruded cut of the first part, thus forming a solid connection. Figure 2.2 shows, similarly as for the first part, the top, front and isometric view of the second part.

Figure 2.2: Top, front, and isometric view of part two of the attachment system The squared-shaped extruded base should in theory have the same dimensions as the extruded cut of the first part (10x10x4 mm), but this turned out to be too large. The dimensions of the extruded base were varied until the two parts could be connected.

The dimensions of the extruded base for which this is the case are 8.6x8.6x3.5 mm.

In figure 2.3 parts one and two of the attachment system are shown on the left and middle respectively. On the right, the connection of the two parts is shown, on top of which the markers will be glued.

(18)

Figure 2.3: Part one on the left, part two in the middle, and the connection of the two parts on the right

2.1.2.2 Markers

The design of the markers will be discussed in two parts: the outside of the marker section and the inside of the marker section.

2.1.2.2.1 Outside of the marker

The cameras in the OptiTrack lab will seek markers with a certain circularity and size [20], which leads to the decision of making spherical markers. Four different size markers will be created for testing of its influence on the accuracy of the biopsy: 8, 12, 16, and 20 mm in diameter. To be able to glue the markers on the attachment system (the right object on figure 2.3) a small portion of the sphere will be removed from the bottom creating a horizontal bottom. This horizontal cut will be made at 2 mm from the bottom of the sphere following the main axis. Figure 2.4 shows the design of the outer-markers in the four different sizes.

Figure 2.4: SolidWorks outer-marker designs for marker sizes 20, 16, 12 and 8 mm (from left to right)

2.1.2.2.2 Inside of the marker

The inner-marker will be filled with petroleum jelly to be visible on the MRI scan. In section 2.1.1 it was described that a 5 mm diameter sphere (’bubble’) should yield a clearly recognisable object in the MR image. It also states that for auto-segmentation of the markers a 2 mm gap between the petroleum jelly and phantom is required, which is already met by the height of the attachment system at which the markers will be glued.

The diameter of the inner-marker is lower-bounded by the 5 mm diameter require- ment and upper-bounded by the diameter of the outer-marker, which is 8 mm for the smallest variant. For the 8 mm outer-marker diameter this leads to the choice of an

(19)

inner-marker diameter of 7 mm. For all other marker sizes, the same inner-marker diameter size is used to prevent that this parameter possibly influences the results of the experiments. Figure 2.5 shows the markers from figure 2.4 in the same order but under a different angle, making the inner markers visible.

Figure 2.5: Markers from figure 2.4 under a different angle to show the inner markers

2.1.3 Petroleum jelly

The inner markers of figure 2.5 will have to be filled with petroleum jelly to be visible on the MRI scan. To put the petroleum jelly inside the markers, the following procedure will be used:

• Put the petroleum jelly inside a syringe

• Heat the syringe in a bain-marie

• Poor the liquid petroleum jelly inside the marker

• Clean the marker (to prevent any petroleum jelly coming on the outside of the marker)

The petroleum jelly reaches a temperature at which it becomes a liquid, but not too hot for it to melt part of the marker. When the liquefied petroleum jelly is poured inside the marker, it cools down and solidifies, creating the required petroleum jelly sphere.

2.1.4 Quantity

The total amount of markers that can be placed on the phantom depends on the size of the phantom and the desired spacing between markers. If the spacing is too small, too many markers will be on the phantom, which might result in marker overlap (elaborated in section 2.5.2) and too little space for needle insertion. However, when the spacing is too big, only a small amount of markers might be put on the phantom. Considering the aforementioned requirements a spacing of about 3 cm between individual markers has been chosen. Based on the size of the phantom that will be used in this assignment, it is possible to put 14 markers on the phantom.

(20)

2.1.5 Retro-reflective tape

To make sure that the OptiTrack cameras can capture the location of the markers, retro-reflective tape will be put on the markers. The tape is put on manually by first putting a strip of tape around the entire marker at its base and folding it on the rounding of the sphere. Subsequently, a circular piece of tape is put at the top and folded down, connecting with the strip of tape. In figure 2.6 multiple markers with tape on them can be seen.

Figure 2.6: Side view of the phantom with all 14 markers on it

(21)

2.2 MRI processing

This section will discuss the required MRI processing of the phantom data.

2.2.1 MRI scan

The phantom in figure 2.6 with a maximum of 14 markers is placed inside the MRI scanner of the MRI machine present at Technohal at the University of Twente (Esaote G-scan Brio 0.25T). A 3D balanced gradient echo sequence is performed to acquire the MRI scan. This acquisition method is used as it renders a clear image of a PVC phantom with only a little noise. In the software (E-MRI) the following settings have been chosen to acquire a clear scan of the breast. Table 2.1 shows the settings, where the starred values indicate whether a box is unchecked or checked.

Table 2.1: Settings used to create the MRI scan

Setting Value Setting Value

EchoTime 5 Distance 0

FlipAngle 60 3DEncodingFov 210

RepTime 10 3DPhases 92

NumAcq 1 (%)Clip 3D phases 36.72

SeriesNum 3 ImageFov 210

ReadingFov 210 MagnetFov 270

EncodingFov 210 ImageDim 256x256

SampleNr 180 EncodingDir Rows

Phases 180 ImageRes 0.82

%0vrsmp 110 (unchecked*) EncodingRes 1.17 (0.82)

%0vrsmp 3D 100 (unchecked*) QualityFactor 120 Read Enc. Inv. unchecked* ReadingRes 1.17 (0.82)

Hamming Filter Low ScanTime 05:38

IsotropicAcq checked* 3DEncodingRes 2.28 (0.82)

Packs 1 ScanNumber 1

Current Pack 1 SAR (W/kg) 0.028 (Partial body)

The software returns DICOM files which combined show the MRI scan of the entire breast. These files will be used to locate the markers and lesions in the breast and create a 3D model of the breast.

2.2.2 Analysis

In this subsection the different aspects of the analysis will be dealt with, starting with the coordinate system that will be used.

2.2.2.1 Coordinate system

From the combined DICOM files, the locations of the markers and lesions can be retrieved. For this assignment, it does not matter in which coordinate system the location of the lesions and markers are represented, as long as the same coordinate system is used for both. This is because a transformation will be performed between

(22)

the phantom in the imaging phase and the phantom during the biopsy procedure.

This transformation is used to project the phantom in its current position, assuming that the markers are identified correctly (elaborated more in chapter 3).

An important coordinate system that often is used is the patient coordinate system (anatomical space) in which the following three planes are defined: axial, coronal, and sagittal. Multiple bases in this coordinate system can be defined, depending on the direction in a certain plane. In DICOM images often the LPS base is used [24]:

LP S =

from right towards left from anterior towards posterior from inferior towards superior

For the analysis of the MRI scan made of the phantom in figure 2.6 the LPS base will be used.

2.2.2.2 Markers and 3D model of the breast

To locate the markers in the MRI scan, software from Vincent Groenhuis is used, which returns both the 3D model of the breast (faces and vertices of the phantom) and the centroids of the markers in LPS coordinates among other parameters. The 3D model of the breast, as viewed in Ultimaker Cura, in a prone position can be seen in figure 2.7. The centroids of the 14 markers on the phantom during imaging in LPS coordinates are stated in table 2.2, with the center of the LPS coordinate system being located around the center of the phantom. Some of the markers are also shown in the figure. The red parts on top of the phantom in the figure are only there to indicate that the phantom can not be printed in this configuration.

Figure 2.7: 3D model of the phantom in the prone position showing the location of the center of the phantom and some of the markers.

(23)

Table 2.2: LPS coordinates of the 14 markers on the phantom during imaging Marker L (mm) P (mm) S (mm) Marker L (mm) P (mm) S (mm)

1 -22.57 -48.73 -21.28 8 -43.55 6.45 -29.63

2 -6.93 -21.41 -39.69 9 -54.50 -21.28 9.37

3 -21.56 37.67 -44.67 10 47.65 11.77 7.11

4 -0.07 16.01 58.97 11 29.03 -6.31 38.04

5 36.86 -33.79 -8.67 12 -6.03 -53.79 24.55

6 -66.05 24.48 6.60 13 -25.25 -17.07 47.07

7 25.57 21.54 -44.04 14 -51.05 36.69 55.31

2.2.2.3 Lesions

To acquire the centroid locations of the lesions the program Slicer (version 4.10.2) is used. The DICOM files are loaded in the program, after which slices of the phantom are depicted for certain values of R, A, and S. Slicer does not work with LPS but with RAS, which requires some minor alterations after the lesions have been found. The general plan for finding lesions is by scrolling through the different layers and looking for larger grey spots that might resemble lesions (which can be checked visually with the phantom).

In figure 2.8 a screen capture is shown of Slicer showing a slice of the phantom in three different planes. In the top right, a 3D view of the breast can be created, which is not shown in the figure. The MRI scan is analyzed and fiducials are placed on the locations of the lesions (one fiducial is shown in figure 2.8 in red). The RAS coordinates of these fiducials can then be accessed in Markups and can be changed to LPS by setting L = −R and P = −A, giving the results in table 2.3.

Figure 2.8: Screen capture of a phantom slice in the three planes and 3D view on top right

(24)

Table 2.3: Coordinates of the lesions in RAS and LPS frames

Lesion R (mm) A (mm) S (mm) L (mm) P (mm) S (mm)

1 -5.12 3.69 -1.27 5.12 -3.69 -1.27

2 26.60 -9.02 33.33 -26.60 9.02 33.33

3 32.57 6.15 9.43 -32.57 -6.15 9.43

4 23.24 33.82 14.11 -23.24 -33.82 14.11

(25)

2.3 Needle-steering guide

In this section, the information will be provided regarding the needle-steering guide that will be used to guide the biopsy needle. This needle-steering guide has been designed and built by Vincent Groenhuis along with the code that is used for the actual steering of the needle. There will also be some explanation regarding the calculation of the needle tip position and how this has been validated. The needle- steering guide (Aimbot device) will be used to steer the needle towards the lesion and the markers that are placed on it will be used to accurately determine the location of the needle tip. Figure 2.9 shows the design of the Aimbot device, consisting of two main parts: the base and the needle holder.

Figure 2.9: Needle-steering guide [5]

2.3.1 Base

The base measures 70x84x26 mm and contains the following components:

• an ESP32 feather board

• two DC gearbox motors

• a lithium battery

• a monochrome display

• a toggle switch

• a button switch

(26)

The ESP32 feather board is paired with the computer using a Bluetooth dongle and data is exchanged over Bluetooth Low Energy (BLE). In Arduino IDE, code has been written that changes the servo angles of the two DC gearbox motors to achieve the desired configuration. The device can be switched on using the toggle switch and the autonomous needle-steering is done by pressing the button switch. The monochrome display informs the user of the battery level, whether the Aimbot is linked to the computer and whether the button switch is pressed [5].

2.3.2 Needle-holder

The needle holder consists of two parts, the first being the part on which the four markers of the Aimbot device are placed as can be seen in figure 2.9. The second part connects the first part to the base, actually holds the needle and is able to rotate the needle to the desired orientation.

Figure 2.10: SolidWorks design of the first part of the needle holder (schematic) [5]

Figure 2.10 shows the schematic SolidWorks design of the first part of the needle holder. The locations of the markers have been chosen in such a way that the inter- marker distance with the other markers is unique for all markers, which will make their identification much easier. The height of this part is 3 mm. In the main beam, three screw holes have been designed to connect the two parts.

Three different views of the second part of the needle-holder are shown in figure 2.11, where the first two views clearly show the three screw holes in combination with a bevel gear that is used (in combination with the motors located in the base) to rotate the part in the desired orientation. The hole in the center of the gear is used to ensure the fastness of the connection between base and needle-holder (with a simple

(27)

bolt-nut connection). The isometric view in figure 2.11c shows a hole on top, which is where the needle moves into and passes through the entire part. The dimensions of the rectangular beam of this part are 10×50×6 mm.

(a) Back view (b) Front view (c) Isometric view

Figure 2.11: Three different views of the second part of the needle-holder

2.3.2.1 Biopsy needle

For choosing the biopsy needle two main characteristics have to be taken into account:

needle length and needle diameter. If the needle diameter is too low, needle-tissue deflection might occur, which currently can not be accounted for. The most important thing to consider concerning the needle length is that the entire working space should be accessible, meaning that the needle can reach every position inside the phantom.

Taking this all into consideration the following length and diameter have been chosen:

Table 2.4: Needle characteristics Length (mm) 111 Diameter (mm) 2

The needle passes through the entire needle holder with a length of 50 mm, which means that an effective needle length of 111 − 50 = 61 mm exits the needle holder.

(28)

(a) Phase 1

(b) Phase 2

(c) Phase 3

(d) Phase 4

Figure 2.12: Four different phases of the biopsy needle mechanism

Figure 2.12 shows the biopsy device in the four different phases that it can be in.

The biopsy device begins in phase 1 (figure 2.12a), after which the blue mechanism at the left can be pulled. At some point phase 2 is reached (figure 2.12b) and pulling even further the final extension state, phase 3 (figure 2.12c), is reached. The blue mechanism can be pushed back to its original state which pushes out the middle part of the needle as well. This creates phase 4 (figure 2.12d) where the outer part of the needle is still pulled back a little. When more force is put on the blue mechanism the outer part of the needle will be shot forward and collect the tissue due to its bevel shape.

(29)

2.3.2.2 Obtaining needle-tip position

To obtain the position of the needle tip for a certain marker configuration a rigid transformation has to be performed between the original and the current configura- tion. The problem comes down to minimizing the error function in equation 2.1.

min

R,t||R · A + t − B||2 (2.1)

The error function (equation 2.1) is minimised by finding the optimal rotation R and translation t for data set A to match data set B after applying the transformation. To calculate the rotation R, it is required to eliminate the translation part. The centroid value of the two data sets are calculated and subtracted from their data sets:

Asub= A − mean(A) (2.2)

Bsub= B − mean(B) (2.3)

Since there is no translation anymore between the data sets, it is possible to calculate the rotation between the sets, which will be done using Singular Value Decomposition (SVD). SVD factorises an m × n matrix in three different matrices (an m×m complex unitary matrix, an m×n rectangular diagonal matrix with non-negative real numbers on the diagonal, and an n×n complex unitary matrix) [25]. The Matlab function [U, S, V] = SVD(H) factorises matrix H in those 3 matrices and stores them in U, S and V respectively. In this function, the matrix H will be the cross-covariance matrix of the two data sets.

The rotation R can be found by multiplying the two complex unitary matrices that are calculated using Matlab’s svd function in the following manner:

R = V · UT (2.4)

The calculated rotation is applied to the centroid of data set A, such that both centroids have the same orientation. The translation can then easily be found by subtracting the rotated centroid of set A from the centroid of set B:

R · centroidA+ t = centroidB

t = centroidB− R · centroidA (2.5) The transformation matrix between the current and original configuration can then be defined to be:

TF =R t

~0 1



(2.6) In the transformation matrix TF in equation 2.6, R is a 3 × 3 matrix, t a 3 × 1 column vector and ~0 a 1 × 3 row vector of all zeros [26].

To obtain the needle tip position two coordinates frame will be defined, one at the center of part 1 of the needle holder and one at the needle tip, as can be seen in figure 2.13. The first coordinate system (Coordinate system1 in figure 2.13) is located at the middle of the green rectangular beam (at the surface), resembling the point in

(30)

figure 2.10 where all distances are calculated from. This coordinate system will be denoted as Ψpart1. The second coordinate system (Coordinate system2 in figure 2.13) is located at the needle tip and has the same orientation as Ψpart1. This coordinate system will be indicated as Ψneedle tip.

Figure 2.13: Coordinate systems used on the Aimbot device

The direction of the three axes has been chosen to match the directions of the Opti- Track coordinate frame which will be elaborated on in section 2.5. A rigid transfor- mation will be performed between the marker coordinates in the original frame (figure 2.10) and the marker coordinates returned by the OptiTrack system. The markers in the original frame are stored in the variable Pe:

Pe =

x1 y1 z1 x2 y2 z2

x3 y3 z3

x4 y4 z4

=

30.41 27.53 −6.5

−30.41 27.53 −6.5 40.88 −37.66 −6.5

−51.12 −47.76 −6.5

(2.7)

The first two columns of Pe follow straight from the schematic of figure 2.10. The -6.5 mm in the third column comes from the radius of the Aimbot markers in the negative Z-direction. The Aimbot marker locations in the OptiTrack frame, stored in the variable Pc2, are then used to find the rigid transformation matrix Tpart1

(between original and current Aimbot markers configuration). In Tpart1 both the orientation and location of Ψpart1 are stored. To obtain the location and orientation

(31)

of the needle tip a transformation matrix has to be created that describes the rotation and translation between Ψpart1 and Ψneedle tip, which is then multiplied with Tpart1

to obtain Tneedle tip.

Tneedle tip = Tpart1· Tpart1 to needle (2.8)

Tpart1 to needle=

1 0 0 0

0 1 0 86

0 0 1 6

0 0 0 1

(2.9)

There is no rotation between Ψpart1 and Ψneedle tip which means that the rotation part of the transformation matrix from equation 2.9 is a 3×3 identity matrix. There is a translation between the two coordinate frames: 0; 86; 6. The location of the needle tip equals the translation part from Tneedle tip (equation 2.8).

2.3.2.3 Validation of needle-tip position

The location of the needle tip will be verified by moving the needle tip to one of the markers that are on the phantom and comparing the calculated needle tip position to the marker position. At the time of validating the needle tip position, 6 markers were present on the phantom, which means that the needle tip location will be validated using 6 markers. The locations of the markers that are returned are the locations of the centers, whilst the needle tip will be placed on the edge of the outside of the marker. The radius of the marker will be added in the calculation of the needle tip to compensate for this difference. Table 2.5 shows the results of the needle tip validation. ex, ey, and ezare the differences between compensated needle tip location and marker location in x, y, and z-direction respectively, whereas in the Norm column the euclidean distance is shown.

Table 2.5: Needle tip validation results on 6 markers Marker ex (mm) ey (mm) ez(mm) Norm (mm)

1 1.58 1.33 -0.52 2.13

2 1.22 -3.02 2.65 4.20

3 1.13 -0,50 -2,24 2.56

4 -1.25 -5,00 2.23 5.62

5 -2.42 7.33 1.65 7.89

6 0.37 2.68 -0.22 2.71

The norm for all 6 validations lies between 2.13 and 7.89 mm, with an average norm of 4.2 mm and a standard deviation of 2.2 mm. It has to be noted that the validation experiments are performed by hand and are therefore prone to errors. A setup can be designed in which the needle and Aimbot markers are kept in place and then in theory a perfect validation experiment can be performed. This perfect validation, however, is not required for this master’s assignment and will thus also not be performed.

(32)

2.4 Lesion tracking

In [27] different 3D interpolation techniques are evaluated based on how good the interpolated motion compares with the gold standard of the original segmented pro- jection data. In their work, interpolation is required to obtain a dense motion vector field (MVF) from a sparse MVF. This MVF would then be used to compensate for the motion, overcoming the limitations induced by the FDK algorithm (a 3D image reconstruction algorithm created by Feldkamp, Davis, and Kress [28]). In their work, they evaluate the Thin Plate Spline (TPS) algorithm, Shepard’s method, simple av- eraging, a smoothed weighting function, and the standard FDK algorithm.

For their evaluation they used three different types of data: Phantom, Porcine, and Clinical human data (three different data sets). The quantitative evaluation is performed on the following four parameters: Normalised root mean square error, Universal quality index, Dice similarity coefficient in 2D projection space and Mean contour deviation ε in 2D projection space. For the three different types of data and the different evaluation parameters, TPS turned out to be the best 3D interpolation method, having the best values in each combination of data and evaluation parameters (or tied for first with other interpolation methods).

2.4.1 Validation results

In the research of Lagomarsino, a validation of the location of a lesion on 15 trials for both the TPS and RT algorithms is performed. An NDI tracker was placed inside the phantom, whose location could be retrieved using an NDI system (present at the University of Twente). The estimated location using the two algorithms was compared to the actual location during a biopsy. Results of the validation process for both algorithms indicated that the TPS algorithm produces a better estimation of the lesion location with an average error of 1.16 mm compared to an error of 1.34 mm for the RT algorithm [5]. The outcome of this validation in combination with the outcome of the research performed by Schwemmer et al. [27] leads to believe that TPS is the best interpolation method for this research as well.

2.4.2 Theory of TPS

In the TPS algorithm a thin sheet of metal is considered, which is bent based on a set of landmark points, with the purpose of minimizing the required energy for bending this sheet [29]. To explain the mathematical part of the theory of TPS [30] is used. To clarify the matrices that will be used during the explanation, numerical values will be used, obtained from OptiTrack with 7 markers on the phantom (markers 2, 5, 8, 9, 11, 12, 13 from table 2.2). The TPS method maps coordinates using both linear (affine transformation) and nonlinear (weighting factors) parts. The affine transformation defines the overall shape of the spline whereas the nonlinear part is used to produce deformations [31]. The weighting factors W and affine transformation factors A can be found by solving the following equation:

Referenties

GERELATEERDE DOCUMENTEN