Preliminary Program

Short summary of the lectures

Welcome to the BrainHack!

Our BrainHack chairs and distinguished authorities representing the state of São Paulo, Brazil, and the province of Alberta, Canada, will welcome the participants to the first São Paulo - Alberta: BrainHack.

BrainHack Projects

Description of the three challenges proposed in the BrainHack, the data and computational infra-structure available to be used for developing innovative solutions to the proposed challenges. Formation of multi-disciplinary teams and assignment of a challenge for each team.

BrainHack Mentors' Introduction

Each BrainHack speaker/mentor will have one minute to introduce themselves and their area of expertise. This will be useful for the BrainHack teams when seeking for guidance/mentorship for developing their BrainHack challenges.

Magnetic Resonance Image Reconstruction and Compressed Sensing

In the first portion of the talk presented by Dr. Garrido, basic concepts of magnetic resonance (MR) image acquisition and k-space will be introduced. A particular application of MR images in aging studies will be presented, specially using these images as a brain morphometric tool. In the second half of the talk, Dr. Frayne will discuss the advances in MR acquisition and the importance of developing techniques to reduce MR scan times, especially techniques related to compressed sensing. He will touch on aspects of traditional iterative compressed sensing techniques that usually require a sparsifying transform and newer approaches based on machine learning, specially deep convolutional neural networks.

Image Analysis Methods

In the first portion of the talk presented by Dr. Gobbi, he will cover several image processing topics, such as the DICOM medical image storage standard, batch processing, and labelling using tools like the advanced normalization tools (ANTs - http://stnava.github.io/ANTs/). Following Dr. Gobbi, Dr. Ferrari will describe the current image analysis projects in development in his research group (Biomedical Image Processing group - https://www.bipgroup.dc.ufscar.br/) on computer-aided diagnosis systems for segmenting and classifying Multiple Sclerosis (MS) lesions, and also for detecting and classifying structural changes in MR images that can potentially be used as biomarkers of Alzheimer's disease.

Medical Image Databases

In the first portion of the talk presented by Dr. Salluzzi, she will talk about image anonymization, quality control and PACS (Picture Archiving and Communication System) and the challenges in maintaining and doing quality control of large images databases. She will illustrate her talk with examples that she faces on a daily basis as the manager of the Calgary Image Processing and Analysis Centre. In the second half of the talk, Dr. Traina is going to discuss the concepts and novel techniques aimed at content-based image retrieval (CBIR) for medical databases. The main challenges posed to the image processing and database communities in order to provide the basis to build robust CBIR systems will be discussed.

Machine Learning I

In the first portion of the talk presented by Dr. Brown, he will discuss applications of machine learning to brain imaging, with a focus on clinical applications related to mental health. He will include a discussion of factors for success and failure in this sort of inter-disciplinary research, including fundamentals of proper machine learning methodology, data size and quality, issues of generalizability, and common pitfalls that can produce false positive results. Following Dr. Brown, Dr. Avila will explore Deep Neural Networks for skin cancer classification, and the use of Generative Adversarial Networks for data augmentation. One of the issues on classification problems is the lack of annotated data, which are expensive and require a lot of effort from specialists. We can alleviate this issue by recycling knowledge from models trained on different tasks, in a scheme called transfer learning; and/or by generating more data which consists of applying affine transformations to the original data (data augmentation).

Machine Learning II

In the first portion of the talk presented by Dr. Lotufo, he will present the various convolutional network architectures used for semantic segmentation, with applications in MR imaging with the objective of segmenting the brain and its structures. One of the main difficulties in using machine learning techniques in medical imaging is the creation of annotated data that serves as a gold standard for training. He will discuss the use of silver standards, a novel data annotation technique developed by his group, that combines the results from several traditional segmentation algorithms to create the so called silver standard. In particular, he will discuss how the Calgary- Campinas public brain MR image database was constructed. Following Dr. Lotufo, Dr. Forkert will present his group research on developing and evaluating new image processing methods, algorithms and software tools for the analysis of medical images. This includes the image-based extraction of clinically relevant parameters and biomarkers describing the morphology and function of organs, with special focus to the human brain.

Abstracts Power Pitch

Each participant with an accepted abstract at the BrainHack will have 3 minutes to present a summary of their work and get feedback from the audience. Presenters can also prepare full electronic-posters, i.e. slides, which will be displayed on screens throughout the BrainHack. This will allow them to receive more personalized feedback.

Hands-on Sessions

The hands-on sessions are for the teams to sit together, discuss and start implementing solutions for their BrainHack challenges. This is a team work exercise. For example, people from a physics background may understand better the process of acquiring MR images, but people from a computational background are probably better suited for translating new ideas to efficient software tools. Finally, people with a medical background are the ones who look and interpret these images and they have deeper knowledge about brain anatomy, so they are the ones most qualified to judge the quality of MR images or the segmentation of a brain structure.

At the end of the BrainHack, we expect to see innovative solutions for the proposed challenges, but we also expect that people understand the value of multi-disciplinary research and different points of view when developing new solutions for problems in medical image analysis.

Socializing/Networking

The ratio between number of students and faculty and/or industry members attending the event is relatively low (~5:1) compared to other events. We encourage participants, specially students, to leave their comfort zone and interact with more experienced researchers, investigators from a different discipline. The São Paulo - Alberta: BrainHack is a unique opportunity to grow your network and discover prospective collaboration opportunities. Use our coffee breaks, lunch breaks, and BrainHack dinner to meet new people!