My I underestimated the difficulty and complexity of the

My undergraduate education ensured me about the kind of person I am, and my graduate study
at Johns Hopkins helps me realize the kind of person I truly want to be – a university faculty with
knowledge, warm heart and keen mind in both research and the outside world. Currently, I am a
master student, a teaching assistant, and a research assistant.
I always welcome challenges and treat them as opportunities to improve myself. The most
challenging course I’ve taken so far is Medical Image Analysis taught by Dr. Jerry Prince. There
were 21 students in that class, and about 1/3 of them were PhD students. I noticed that I was
only an average student at the beginning, since I wasreally struggling to understand the materials
in the lectures. On the other hand, I became really excited to be able to work with so many
awesome people. In the course project, I applied the knowledge I had learned from other courses
and my research project. Due to my hard work, in the project evaluation session, our team ranked
top 2 among the 7 teams, and my responsible project ranked top 1. The most joyful moment
occurred when I realized that I really made progress because of my effort.
As a student, never did I settle for an A-. The only A- I got so far during my graduate study was
from Image Processing and Analysis I. However, I neither became frustrated nor simply let it go.
Instead, I carefully rethought about the possible reasons why I didn’t do well in that class, and
realized that I underestimated the difficulty and complexity of the course; some concepts that I
assumed to understand actually had deeper meanings. Then, I decided to do something to get a
second chance – to become a TA of that course. However, the road to happiness is strewn with
setbacks. Usually there are only 2 TAs needed, and the other two applicants received higher
grades in the course than me. I didn’t get instructor’s approval to be a TA by the end of the first
week, but I didn’t give up. I knew I must grasp this chance to better understand the materials and
also improve my teaching skills. I talked to the instructor, Dr. John Goutsias, and told him “I know
I didn’t do a good job compared with the other two applicants, but that’s why I’m here. I can
prove I am able to do a good job.” In the end, an exception was made – he hired me along with
the other 2 applicants.
Since I was an undergraduate student, my main focus has always been on conducting research
about image processing and making innovative designs. I’ve worked at Dr. Chenghui Qian’s 216+
Innovation Design Laboratory for 2.5 years at Jilin University. My team has proposed many fancy
innovative designs during that time. One of them is a supervising system for drivers, which can
detect driver’s fatigue using a camera and give an alert when necessary. The project finally won
the Excellent Project Prize of National College Student Innovation Design and a paper was
published. Though achievements had been made, I realized that I was doing more engineering
than research; I wasn’t creating new things. That’s why I chose Johns Hopkins as the starting point
of my graduate education. At Hopkins, I joined Dr. Jerry Prince’s lab IACL in my second semester.
My research project at IACL is to do quality assurance (QA) for a cerebellar lobule segmentation
software. To be honest, this project was quite challenging for me, since I didn’t have enough
background knowledge at the very beginning. It is the undergraduate lab experience that helped  me to make plans properly without being overwhelmed. I subdivided my QA project into several
implementable steps. To better understand my project, I told myself to first understand the
software code and be able to run it by myself as soon as possible. Then, doing literature research
to get a picture of the method I might propose. The first 2 months were really tough, especially
when I had 3 courses to take at the same time. I took notes every day about the problems I
solved, and new problems to address. Finally, I was quickly able to keep pace with others in the
lab because of my systematic plan making.
My current research interest involves quality assurance on medical images using statistic models,
probabilistic models and machine learning techniques. The difficult days of my personal research
experience occurred this summer when I was a full-time research assistant at IACL. After I tried
several methods and none of them worked well, I realized that I should step back and rethink the
problem I was facing. Along with the help of Dr. Jerry Prince, I noticed that for a long time, I only
focused on finding outliers of the final results, which is an intuitive approach but may not be the
most sufficient. For a given medical image processing software, the intermediate output
including the final output together are sequential data, which inspired me to use hidden Markov
models that I had learned from an audio signal processing course. The result was quite promising
and I got my paper accepted for SPIE 2018. However, the involvement of human when doing QA
and the lack of generality of QA methods always bother me. Therefore, the next stage of my
research project plan is to conduct QA on other medical image analysis software using
probabilistic and statistic models without expert delineations as ground truth. Rather than
focusing on software outputs as before, I am thinking about switching my focus onto software
itself, and conducting QA in a systematic point of view, such as modeling system behaviors based
on input-output pairs.
The QA experience helps me realize that the limited amount of training data and the variation
between datasets from imaging procedures make medical image analysis unique from general
CV. Therefore, I believe, there will be a growing demand of researchers in this field who can
propose accurate and reliable methods, and the focus of machine learning should switch to more
explainable models. The past research experience in machine learning, probability and statistics
ensured me to be able to undertake various medical image processing projects. Amongst all
universities around the world, Yale University has a high reputation in medical imaging. At Yale,
Dr. James Duncan’s research in tissue classification, Dr. Lawrence Staib’s research in medical
image segmentation and registration using deformable models and Dr. Hemant D. Tagare’s
research in image segmentation and outlier detection match my interests well. Further, the
courses I took at Hopkins ECE gave me a good understanding in medical image processing and
data analysis from an engineering point of view. In my PhD study, I plan to take more math and
physiologic based courses to strengthen my knowledge in biomedical engineering research. This
idea also matches well with the core courses requirement of Yale BME.
It is the progress I’ve made that makes me confident of being a successful PhD student. I believe
with the shared interests and the growing enthusiasm in medical image analysis, I can achieve
more academic goals together with my advisor and colleagues at Yale in the future.