Youngho Seo, Professor

Closed (1) Big Data in Radiology Research (e.g., medical deep learning, data science, computer vision, natural language processing, big data organization)

Applications for Fall 2018 are now closed for this project.

UCSF Department of Radiology and Biomedical Imaging is excited to offer a combined educational and research opportunity for motivated undergraduates students in the Big Data in Radiology (BDRad) research team.

UCSF collects over 300,000 studies every year with approximately 10 million total studies, but most data remain unorganized and untapped. Our laboratory specializes in quantitative analysis of images from CT, PET/CT, and PET/MRI using image processing and machine learning. We work in close collaboration with UCSF Institute for Computational Health Sciences, to apply data science and deep learning techniques to uncover imaging biomarkers from clinical radiology studies. We identify features that predict patient survival, therapeutic response, tumor characteristics, seriousness of a particular pathology, and others. Additionally, we analyze radiology text reports, which are extractions of most important information from the radiological images.

Past and current projects include lung nodule characterization from chest CT, integrating survival analysis into convolutional neural network, automated generation of radiological impression using neural machine translation, using deep learning to identify rib fractures from chest x-rays, and automated MRI protocoling using free-text clinical indications, just to name a few.

You will work closely with a resident physician-engineer and data scientist mentor, Dr. Jae Ho Sohn, along with other members of the team from UCSF, UC Berkeley, and MIT. Together, you will report to medical school faculty mentor, Dr. Youngho Seo. We have multiple state-of-art equipment with Pascal Titan X GPUs for deep learning with over 15 terabytes of public, semi-public, and private database as well as modular keras/PyTorch libraries to facilitate deep learning.

Every week on Monday evening (sometimes Tuesday evening), we will conduct seminar series, initially discussing introduction to clinical radiology, computer vision, natural language processing, and scientific manuscript writing. This will be followed by student-led seminars on emerging deep learning techniques and papers. This remains flexible - in the past, we discussed papers, had guest lecturers from academia & industry, and went through FastAI coursework.

Upon successful progress, it is expected that student submit/present at a national research meeting. Students are encouraged to seek out and apply for undergraduate research grants. Many of our students have went on to produce peer-reviewed publications. Pre-health students with interest in machine learning/data science or engineering students with interest in healthcare research or research-heavy biotech industry are especially encouraged to apply.

Dr. Jae Ho Sohn's Profile: https://profiles.ucsf.edu/jae.sohn

Time Commitment (~9-12 hours per week):
- Attend meeting remotely via Zoom on Monday evenings TBD but usually 6:30PM - 8PM (sometimes rescheduled to Tuesday evenings)
- Attend in-person meetings once at the beginning, once at the end of the first semester, and once at the end of second semester.
- Lead 1-3 weekly seminars on deep learning research paper and coding discussions per semester.

Training Opportunity
-Collecting, Annotating, and Analyzing Big Data in Medicine.
-Applying Machine Learning, Statistical Methods, and Image Processing to tackle clinically important problems (specifically, image classification, image segmentation, and word embedding creation using Keras and PyTorch).
-Communicating research problems and results to a variety of audiences including clinicians, engineers, entrepreneurs, and general public.
- Submit a medical deep learning abstract to a national conference such as RSNA with us.
- Prepare a medical deep learning manuscript to be submitted to a peer-reviewed journal

Day-to-day supervisor for this project: Jae Ho Sohn, Post-Doc

Qualifications: Requirement: - Working Knowledge of a programming language (Python preferred, but significant experience in others acceptable). At least one college level coursework in programming. - Working Knowledge of statistics (mean, variance, linear regression, t-test, anova, hypothesis testing). At least one college level coursework in statistics. - Working knowledge of basic machine learning (cost function, cross-validation, overfitting, error analysis). It is expected that students are at least concurrently enrolled in a college-level machine learning course at the time of joining the group. By end of first semester, students must become familiar with CS 189/289A or equivalent (or Stanford 231n). Recommended: - Working knowledge of Tensorflow/Keras or Pytorch. - Working knowledge of Version Control (such as Github). Due to limited compute resources and mentor time, we will be conducting interviews to select the most suitable candidates. Near the end of summer vacation / beginning of fall semester, we will contact you so you get a chance to personally talk with several of our excellent student researchers. We are more likely to find you suitable, if you meet many of the above required/recommended criteria, have proven track record of coding experience in deep learning, GPA above 3.5, demonstrated interest in both engineering and healthcare, and/or scientific paper writing experience.

Weekly Hours: 9-11 hrs

Off-Campus Research Site: UCSF Center for Molecular and Functional Imaging
185 Berry Street, Suite 350
San Francisco, CA 94107
Related website: http://www.radiology.ucsf.edu/research/labs/quantitative-image-processing

Closed (2) Radionuclide molecular imaging using small laboratory animals

Applications for Fall 2018 are now closed for this project.

Our laboratory (UCSF Physics Research Laboratory), in collaboration with other colleagues and groups in UCSF Department of Radiology and Biomedical Imaging, performs research in small animal imaging using dedicated small animal PET/CT and SPECT/CT, important translational molecular imaging modalities. We currently perform oncologic, cardiovascular, and brain imaging research in collaboration with many UCSF and UC, Berkeley faculty members and their associated research groups. Students will have opportunities in assisting to acquire the imaging data and processing, and learning important biomedical research procedures.

Training opportunities:
Our group (UCSF Physics Research Laboratory) provides training to students at all levels, and in particular, clinical relevance of the research projects will be discussed at our regular group meetings for each project that is open for training opportunities. We have many collaborating research groups at UCSF, UC Berkeley, and Lawrence Berkeley National Laboratory. Hence, the training for the students could be tailored to each individual's research interest.

Data acquisition, quality control, data processing, experience with laboratory animals

Qualifications: Required: Basic knowledge of medical imaging and biology Required: Solid time commitment (1-2 afternoons per week minimum)

Weekly Hours: 9-11 hrs

Off-Campus Research Site: UCSF Physics Research Laboratory
185 Berry Street, Suite 350
San Francisco, CA 94107

Related website: http://www.radiology.ucsf.edu/physics
Related website: https://radiology.ucsf.edu/research/core-services/mspc