Frugal Deep Learning for Multi-Resolution Image Analysis
Team work: Marie Reinbigler presented "Frugal Deep Learning for Multi-Resolution Image Analysis" at 1D19 the 29/9/2023 at 10h00.
In the biological field, analysis of histological images, i.e. slices of organs observed by microscopy, is key to improving our understanding of diseases. Uncompressed histological images can reach sizes up to 40GB, which makes the analysis of the whole slide image, manual or automated, challenging.
Our goal is to perform automated deep-learning-based analysis of histological images on inexpensive local hardware with modest computational resources, thus enabling the access democratization to analytic tools while preserving privacy. The core of our approach is to build a scaling processing architecture composed of a pair of differentiable blocks. First, an image-based block will extract slice features into related components. In a second step, a graph-based approach, built on the set of previously detected components, will be deployed to perform the classification. We strive to maximize limited resources utilization by developing a dynamic scheduling strategy leveraging application-specific characteristics in a decentralized setting.
The approach, studied for a specific biological case, can be generalized to any multi-resolution image, e.g satellite or geographical images.