Commentary and Perspective (Invited)
Using interactive deep learning to track cells: A report on a 3-day hands-on training program at IUPAB 2024
2024 Volume 21 Issue Supplemental2 Article ID: e212010
Details
2024 Volume 21 Issue Supplemental2 Article ID: e212010
The rapid advancements in biological imaging and data processing necessitate continuous skill enhancement and practical training. To address this need, a three-day hands-on training program aimed at early-career researchers in biophysics was organised preceding the IUPAB Kyoto 2024 conference.
The primary objectives of the training program were to enhance participants’ practical skills in using the trans-scope AMATERAS [1] and analysing collected data with the ELEPHANT [2] 3D cell tracking platform. Hands-on experience with the microscope and the software allowed us to experience the whole process. We connected through teamwork and networking opportunities, collaboration, and knowledge sharing with other international participants, which enabled us to build our network before attending the actual conference.
The training involved a series of lectures, laboratory exercises, data analysis, and group activities to provide us with comprehensive and practical experience. We attended the hands-on training program A, which was held in Osaka and Kobe before the 21st IUPAB and BSJ Joint Congress 2024 in Kyoto.
The hands-on program had two participant groups: one focused on data collection and data analysis, and another group focused on data collection only. During the training, both groups were mixed so we could work together and share our knowledge.
Samara and Lissy are second-year PhD students at the University of Technology Sydney, and it was our first opportunity to attend an international conference. Our research work focuses on the biophysics of cell membrane models and lipid bilayers. We were both very excited to attend the hands-on training as it gave us a unique opportunity to broaden our horizons and learn about imaging with live cells and organisms.
We arrived in Tokyo and had the pleasure of experiencing a Shinkansen ride past Mt Fuji and the beautiful Japanese countryside to Osaka. As part of the data analysis group, we arrived at the Osaka University campus on 18 June for our first day. We met the organisers of the training as well as the other participants in our group, which included a variety of international and domestic PhD and Master’s students.
To kick off the training, we had a lecture by Professor Takeharu Nagai about his career and the AMATERAS microscope the other group had already been working with. After a short break, we met the other group and had our first lecture on the deep learning cell tracking software ELEPHANT by Dr Ko Sugawara [2] (See Fig. 1).
The conclusion of the day was a visit to Professor Nagai’s lab, the AMATERAS microscope (See Fig. 2), and the current data collection process. The other group had prepared live quail embryos for imaging with the microscope. At less than 24 hours old, they measured less than a fingernail in size, but by looking through a tabletop microscope, it was possible to see the heartbeat of the embryo. It was fascinating to see the sample preparation and how movement is visible. We were shown multiple versions of the AMATERAS microscope, its development, and capabilities. Further, we were introduced to Professor Nagai’s research with bioluminescent plants.
After the program finished, we decided to have dinner together at the university food court. It was our first time at a Japanese university food court, and we were intrigued by the differences between Japan and Australia. The food was delicious and the variety extensive, which left us envious that dinner was provided in food halls as well, whereas at our university there are only lunch services. After an exciting first day, we were pleased that the organisers had reserved rooms for all of us at the Kasugaoka House next to the campus for easy access and a comfortable rest for the night.
On our second day, it was time to make our way to Kobe. Thanks to the guidance of the domestic participants, who guided us expertly, we managed the trip without issues. We rode the Osaka monorail for beautiful views, multiple trains, and finally the Port Liner to the RIKEN campus.
Upon arriving, we had another lecture by Professor Onami and received a practical, hands-on introduction to using the ELEPHANT software. Our lunch break was another great opportunity to connect with researchers, both participants and organisers. The organisers showed us around the vicinity of the campus, and we all selected our own Japanese Bento to bring back to our lunchroom. The break created a comfortable environment where we could exchange insights about our research and career goals, differences and commonalities in research work in different countries, and get to know each other better in a casual setting.
We used the time to work with example data sets to become acquainted with the program and the data processing. As it was our first time working with a program for image processing, it was a steep learning curve, but with the guidance provided, we were able to progress quickly and smoothly. To make the learning more collaborative and foster our teamwork abilities, we worked in teams of two: one participant from the data collection group and one from the analysis group. Together, it was easier to solve problems and figure out the workings of the software as we could assist each other.
By the end of the day, we were able to mark cells, train the deep learning system to identify further cells, and track cell movement over the duration of the data collection available in the file. It was amazing to see how much cells moved and changed shape over time. Although we knew this was the case, it was spectacular to see and work with an actual visualisation.
In preparation for the next day, we had to choose the project we wanted to work on in our group of two. We would have the following day to work with one of three data sets collected previously by the AMATERAS microscope and use our newly acquired skills to analyse the images using ELEPHANT. Once complete, we would then prepare a presentation introducing our data set, analysis process, and results to the training group in the afternoon of the next day.
After this, the organisers offered to take us to a restaurant for dinner to experience traditional Japanese food. We all went back to Kobe together and visited an Izakaya, where we shared plates of sushi, sashimi, chicken karaage, grilled ray fins, goya chanpuru, and matcha ice cream. It was a delicious excursion and lots of fun to share traditional meals. It was then time to return to our hotels and get some well-deserved rest for the next day.
Our third day started bright and early, and we returned to the RIKEN campus once more to work on our first real data analysis project! It is safe to say that we were all equally excited and nervous about whether we would be able to apply what we had just learned and whether we could analyse the data appropriately. Everyone got to work straight away and did their best to re-create the processes of cell identification and definition, training the deep learning system, assessing the quality of the output, making changes to correct errors, and training the system again until cells had been defined correctly. Yesterday’s training was done on a 2D data set, but today some of us ventured into the 3D space and had to learn the navigation impacts when another dimension is added. Once our networks had been trained, they were then able to process larger image areas by themselves and repeat the processes for the same area throughout the timeline recorded in the data set. We could then use this data to label individual cells and track their movement over the duration of the recording (See Fig. 3). As we had limited time and some of our data sets were very large, we decided to utilise our lunch break to let the software compute our data sets.
Upon our return from lunch, we saw the results of our deep learning network setup and whether the computation was successful. We were excited to see the outcome, but it was at this point we learned that it is crucial to select the area of interest carefully in case the entire image area cannot be analysed simultaneously. For a beginner in image processing, this was very valuable insight, and we can apply this to our future work even outside of this application. Once we had gathered our insights, we worked on our presentation. We used the opportunity to introduce ourselves and our institutions, adding information about the AMATERAS [1] microscope, the ELEPHANT [2] software, our dataset, our data analysis process, and our results.
After we had finished our data analysis and presentation preparation, the team at RIKEN organised a tour through a variety of laboratories and introduced us to the work done at the institute. We saw a multitude of microscopes and microscopy setups currently used in research. We were also taken to a lab that utilised robots to help with physical tasks. It was impressive to see the work that the Maholo LabDroid can do, and it was mesmerising to watch a robot completing tasks with which we are so familiar.
For our final presentations (See Fig. 4, 5) of the hands-on training program, we all gathered in a lecture room, and Professor Nagai and Professor Onami rejoined us to listen to our findings and experiences. As every group had worked on a data set in their own way, it was great to hear about their processes and findings, the difficulties they faced, and how they handled them. This was a great opportunity to learn from each other and gain insight into different thought processes and methodologies.
To conclude the training, the RIKEN institute organised a farewell party for us. We were able to gather once more and reflect on our experiences with the program as well as our hopes for the upcoming IUPAB conference. We enjoyed some Japanese food and snacks and also tried some local sake to celebrate a successful training.
The hands-on training program was a once-in-a-lifetime experience. We learned a lot by utilising highly advanced and sophisticated cell imaging and analysis techniques, made connections with young and established researchers across the globe, and glimpsed amazing research developments and equipment. While our current PhD work does not include any work with live cells, we are both excited to explore this area in our future. This training was the perfect opportunity to branch outside of our biophysical work with bilayer lipid membranes for a biosensor and drug synergy studies. We gained valuable experiences and insights into microscopy and machine learning that inspired us to seek research positions in this field. We will keep the possibilities of AMATERAS in mind so we can find links to our existing area of expertise and explore the impact on cell membranes and cells in a new context. We hope to work with live cell imaging coupled with the ELEPHANT technology alongside the pioneers of this field, Professor Nagai and Professor Onami, in the future.
We want to thank the organisers for their efforts, their warm welcome to the institutions, the great program they set up for us, and the expert guidance we received during the training. We look forward to meeting again and are hopeful that there will be opportunities for collaboration soon.
