Analysis of Deep Learning Techniques for Surgical Skill Assessment

Desklib provides past papers and solved assignments for students. This literature review explores deep learning applications in surgical skill evaluation.

Review Based Project Literature Review (Secondary Research)
Student's Name
CSU ID
Shekh Risfan Mahmmadrafik
11644125
Project Type Review Based Project
Project Name Deep Learning for skill Evaluation in Surgical Training
Technology Deep Learning
Techniques Skill Evaluation
Domain Surgical Training
1

Version 1.0 _ Week 1 (5 Journal Papers from CSU Library)
1
Reference in APA format that
will be in 'Reference List'
(This give the Reference of the
Journal Paper you are working
on it)
Ziheng W, (2018). Deep learning with convolutional neural network for
objective skill evaluation in robot-assisted surgery. INTERNATIONAL JOURNAL
OF COMPUTER ASSISTED RADIOLOGY AND SURGERY, Volume 13, 1959–1970.
Citation that will be in the
content
(Ziheng, 2018)
URL of the Reference Level of Journal (Q1, Q2, …Qn) Keywords in this Reference
https://doi.org/10.1007/s11548-
018-1860-1
Q2
Surgical robotics, Surgical skill evaluation,
Motion Analysis, Deep learning, neural
network
The Name of the Current
Solution (Technique/ Method/
Scheme/ Algorithm/ Model/
Tool/ Framework/ ... etc )
The Goal (Objective) of this
Solution & What is the Problem
that need to be solved
What are the components of it?
Technique/Algorithm name:
JHU-ISI Gesture and Skill
Assessment Working Set
(JIGSAWS).
Problem: The Current work requires
converting robotic machine
kinematics into gestures similar as
human but it is very expensive to
achieve.
There are the Standard tasks.
Suturing (SU)
Knottying (KT)
Needle-passing(NP)
2

Tools:
Twomastertoolmanipulators
(MTMs) on left and right sides, two
patient-side slave manipulators
(PSMs), and an endoscopic camera
arm.
Goal: The goal of this method is to
achieve high accuracy in machine
gesture with low cost.
The Process (Mechanism) of this Work; The process steps of the Technique/system
Process Steps Advantage (Purpose of this step) Disadvantage
(Limitation/Challenge)
1 Experiment setup: it is the process
of setting required equipments to
solve the identified problem by
collecting data.
In this step the Data comes from the
JHU-ISI Gesture and Skill Assessment
Working Set (JIGSAWS) Dataset.
Time- consuming process. In this,
dataset of 8 different surgeons with
different experiences is involved.
Varying level of experience may
lead to conflicts.
2 Data processing: this is the process
of translating data into usable form
that can be used to withdraw
conclusions.
In this step the data are inputted in
the algorithm named Sliding-window
Cropping Algorithm.
Overfitting is seen when the data is
limited, hence deep learning is
affected.
3 Training: in this process the best
alternative for system development
for deep learning is tested.
In this step the data is validated
using 2 Techniques namely Leave-
one-super trial-out (LOSO) and Hold-
out
Time consuming process as each
step is repeated five times and a
then a trial is randomly selected.
4 Evaluation: it is the process of
comparing to evaluate the
In this step the results are classified
for self proclaimed skill classification
Confusion matrices, if not
evaluated properly leads to wrong
3

performance measures. and globalratingscale(GRS)-based
skill classification using the JIGSAWS
dataset.
predictions.
Lack efficiency.
Validation Criteria (Measurement Criteria)
Dependent Variable Independent Variable
Multivariate time series (MTS) expertise levels of trainees
Valuable information Operators’ skills and proficiencies.
Skill Classification GRS(Global Rating Scale)
Keras library python
Self-proclaimed skills labels Skills levels
Evaluation LOSO(Leave one super-trial out) data
Input and Output Critical Thinking: Feature of this
work, and Why (Justify)
Critical Thinking: Limitations of the
research current solution, and Why
(Justify)
Input (Data) Output (View)
Multivariate time
series (MTS): the
series is
processed
through
automatic
learning.
Multivariate
Experience of
Trainee.
Classification is
directly seen as
output.
Discriminative
assessment of
The Feature of this work is that:
1. It minimizes the Cost of the
Process of Machine Gesture.
2. It provides high Accuracy in all
three tasks.
The Limitation of this Process is that it
heavily depends on the data in JIGSAW
Dataset which is very small.
4

series is recorded
from robot-end
effectors.
surgical skills through
deep learning
architecture.
(Describe the research/current
solution)
Evaluation Criteria How this research/current solution
is valuable for your project
What is the Future work that set by
the author in Conclusion and Future
work section
The Current Solution by the Ziheng is
the creation or development of a large
dataset similar to JIGSAW so that the
problem of expensive machine gesture
similar to that of humans can be
achieved at a low cost.
The average accuracy- ratio between the
sum of correct predictions and total
number of predictions.
Precision-ratio of correct positive
predictions.
This research is important for my future
because it increases my level of
thinking and makes my scope wider.
The interpretation of data is limited due
to black box nature of learning models.
In future the hierarchical representation
would be used to understand the hidden
patterns and take decisions as a deep
learner.
Diagram/Flowchart
5

Figure 1: An end-to-end framework for online skill assessment in robot-assisted minimally invasive surgery.
Reference in APA format that will be in 'Reference List'
(This give the Reference of the Journal Paper that the author selected and improve it (State of art of his work))
6

Fard MJ, Ameri S, Darin Ellis R, Chinnam RB, Pandya AK, Klein MD (2018) Automated robot-assisted surgical skill
evaluation: predictive analytics approach. Int J Med Rob Comput Assist Surg 14(1):e1850.
Dockter RL, Lendvay TS, Sweet RM, Kowalewski TM (2017) The minimally acceptable classification criterion for
surgical skill: intent vectors and separability of raw motion data. Int J Comput Assist Radiol Surg 12(7):1151–1159
Citation that will be in the content
(Fard et al. 2018)
(Dockter et al. 2017)
7

2
Reference in APA format that
will be in 'Reference List'
(This give the Reference of the
Journal Paper you are working
on it)
Kassahun Y, Yu B, Tibebu AT, Stoyanov D, Giannarou S, Metzen JH, Vander Poorten E
(2016) l actions. Surgical robotics beyond enhanced dexterity instrumentation: a survey
of machine learning techniques and their role in intelligent and autonomous surgica Int J
Comput Assist Radiol Surg 11(4):553–568
Citation that will be in the
content
(Kassahun, 2016)
URL of the Reference Level of Journal (Q1, Q2, …Qn) Keywords in this Reference
https://rdcu.be/bqnNk Q2 Surgical robotics Skill learning Skill analysis
Learning to perceive
The Name of the Current
Solution (Technique/ Method/
Scheme/ Algorithm/ Model/
Tool/ Framework/ ... etc )
The Goal (Objective) of this
Solution & What is the Problem
that need to be solved
What are the components of it?
Technique/Algorithm name:
Machine learning (ML)
Tools:
Imperial College Surgical Assessment
Device (ICSAD)
the Advanced Dundee Psychomotor Tester
(ADEPT)
Problem:
In the operation theatre the doctors
have the pressure to take decision on
their own. This creates a mental load
on the operating team.
Goal:
The goal is to apply the Machine
learning to Surgery so in such
situations in the operation room
Sensors
actuators (end effectors)
control architecture
8

Applied Area:
diagnosis and medical image computing
machine can take the decision on its
own.
The Process (Mechanism) of this Work; The process steps of the Technique/system
Process Steps Advantage (Purpose of this step) Disadvantage
(Limitation/Challenge)
1 Automation of the surgical operation: it is
process in which problems are identified and
decision making process is made machine
dependent.
It will speed up the process and save
critical time of patient in the
Operation room.
Requirement of advanced and
expensive machinery.
2 Saving the best strategies of an experienced
surgeon.
If a surgeon is experiences in some
surgeries then we can save the
strategies in the machine to make
the machine work as a pro.
Adverse events can be expected if
data is not fed with precision.
3 Skills learning by the robot: the robot
can be implemented with the skills
either by evaluating the
appropriateness of actions or by
observation of the experiments
conducted by experts.
Experiments are conducted as an
observation activity for the robot
learning. The performance of the
robot is also judged by observing the
actions performed by it.
Time consuming process that
involves huge amount of efforts for
a single learning.
4 Error Analysis: this process is used to
measure the mistakes in the above
steps.
Skills is evaluated through this
technique.
Repetitive and Expensive process.
5 Interaction between surgeons and Safety and compatibility can be Adaptation of the technology by
9

surgical robots. This process is used to
evaluate the compatibility between
surgeons and robots to ensure the
comfort level and safety of machine
learning in operation theatre.
ensured by this process. the surgeons may be time
consuming.
Validation Criteria (Measurement Criteria)
Dependent Variable Independent Variable
Density catheter shape, touching states, entrance and tip points of the catheter.
Average reference surgery Dynamic Time Warping(DTW)
Critical events Knowledge, information, guidance, workflow.
Surgical skills Experience of surgeons
Q-learning Value functions.
Input and Output Critical Thinking: Feature of this
work, and Why (Justify)
Critical Thinking: Limitations of the
research current solution, and Why
(Justify)
Input (Data) Output (View)
Probability of
Density: joint
probability
density
representing
datasets is used
as input to
Catheter-shape,
touching-states,
entrance-and tip
points of the
catheter.
Because this process will put a full stop
on the work-load on the Operating team
and Machines can take accurate decision
in no time by using the Live situation in
the Operation Room.
The Limitation of this Machine Learning
is that It is Time Consuming in some
processes like for inputting the
strategies and It also doesn’t provide
the information about how well the
surgery was performed.
10

predict the
shape, touching
states etc.
(Describe the research/current
solution)
Evaluation Criteria How this research/current solution
is valuable for your project
What is the Future work that set by
the author in Conclusion and Future
work section
The research focuses on decreasing the
work load of doctors so that they can
focus on work independently and
freely.
Training
Competence
Rated checklist based on phantom
bench-top model.
The Importance of this research to my
project is that it highlights that
combining machine power with man
power can result in quick decision
making.
Progress is needed
w.r.t. modeling and real-time
evaluation of deformable
anatomies that would be
accomplished in the future
research.
Diagram/Flowchart
11

Figure 2: Overview of a learning system in surgical robotics
References
Kantor GS (2015) Intravenous catheter complications. http://www.netwellness.org/healthtopics/anesthesiology/
ivcomplications.cfm. [Online; Accessed 16 Sep 2015]
Kehoe B, Kahn G, Mahler J, Kim J, Lee A, Lee A, Nakagawa K, Patil S, Boyd WD, Abbeel P, Goldberg K (2014)
Autonomous multilateral debridement with the raven surgical robot. In: Proceedings of the IEEE international
conference on robotics and automation (ICRA)
Citation
(Kantor al 2015)
12