My tryst with Japan began with an internship in Tokyo in my third year. For two months, I experienced firsthand, the infamous Japanese work ethic, cuisine and culture. By the end of the term, I had decided that I would pursue a career in Japan. In this regard, my senior who was working in Japan suggested ASIA to JAPAN’s program.
Upon reading the program, I found it to be too good to be true! An all-expenses paid trip to Japan just to be interviewed by the companies seemed like an amazing experience. The best part was that the application was completely free! Therefore, I immediately applied. However, I did not get selected in the batch I applied for.
Fortunately, this program happens every month and they keep my application for consideration for each iteration with no extra effort from my side. The following month, I got selected! A total of 3 companies have to consider one to be interviewed to be selected. In the time between the selection to the trip, ASIA to JAPAN staff helped prepare me for the interview. The topics covered were general interview tips, Japanese etiquette, and detailed information on the companies I am interviewing for.
These sessions were very insightful and comprehensive. The staff were very friendly and all my questions, no matter how many, were humored. Once I reached Japan, the entire trip was hassle-free. Transportation and accommodation had been taken care of. We even received portable mobile hotspots. The staff also gave last minute guidance and made sure we were prepared well for the interview. There was also a lot of free time via which I was able to sneak a quick trip to Tokyo.
On the last day, ASIA to JAPAN hosted a party which all of us were able to network well. After being selected by the company, the ASIA to JAPAN staff helped facilitate the salary negotiation and logistics with the company. I have never done such negotiations before and I did not know how to do them either. The experience of the staff helped me get a good compensation. I also do not know Japanese.
Therefore, by request of the hiring company, ASIA to JAPAN set up a Japanese language class up till my joining date. I am currently attending this class and plan to maximize this opportunity to become as proficient in Japanese as I can. Overall, I would highly recommend this program for anybody who’s looking to work in Japan. Since it’s free of cost, there is nothing to lose by just applying. The application is also short and simple and therefore wouldn’t take much time to complete.
My final year project is in the field of medical image analysis. The title of the project is “”A Generalized Deep Learning Framework for Whole-Slide Image Segmentation and Analysis””. Below is a summary of the project. Kindly refer to our paper for more detail (https://arxiv.org/abs/2001.00258).
To diagnose liver cancer, a tissue is first taken from the affected organ and put on a glass slide for observation. A pathologist then observes the tissue and tries to find how much the cancer has developed. He/she then makes a report and submits to the doctor who then decides on the treatment.
However, based on a study conducted by the National Cancer Institute in the USA, it has been
found that pathologists disagree with each other on a diagnosis 27% of the time on an average.
This is due to the fact that histopathological analysis is a complex subject and requires several
years of experience and study to perform accurately. Another reason is that pathologists might
have to look at a large number of cases in a short period of time thereby giving them less time for analysis per image leading to an incorrect diagnosis. This is a serious problem that might lead to mistreatment of the patient.
My final year research project aims to solve this problem. I am developing an algorithm that can
automatically detect cancer from scanned images of pathology slides. This can assist the
pathologist in making an accurate diagnosis faster. Specifically, I am trying to segment
the cancerous region from the whole slide images. Whole slide images are scanned images of the entire tissue sample put on a glass slide (typically 50,000×50,000 pixels).
To accomplish this task, I used deep learning methods. I made this decision because for general segmentation problems (ImageNet), deep learning has given the best results. Specifically, I used an encoder-decoder based fully convolutional architectures. Instead of just one model, I use an ensemble of 3 different architectures to get a better segmentation. Additionally, the input image is very large and the entire image cannot be passed through the network due to memory constraints. Therefore, I have come up with a segmentation method optimized for large images.
I tested the algorithm on the PAIP 2019 Liver Cancer segmentation and got an accuracy of 80% on test data with a sensitivity close to 1. I also participated in a segmentation challenge part of the MICCAI 2019 conference. The algorithm placed second from amongst 800+ participants worldwide and presented our work at the conference (Shenzhen, China). Having validated the efficacy of the algorithm, I developed a GUI on top of the algorithm as a prototype that can be used by pathologists to run the analysis. I’ve hosted the application at digipathai.tech.
