Search papers, labs, and topics across Lattice.
The paper introduces a two-stage Bayesian Optimization (BO) pipeline, 3D-DAP, to automate the selection and parameter tuning of deep learning models for 3D biomedical image segmentation and classification. The first stage uses a domain-adapted benchmark dataset and a novel segmentation quality metric to optimize segmentation models and postprocessing parameters, while the second stage optimizes classifier design choices, including architectures and pretraining strategies, using an assisted class-annotation workflow. The pipeline demonstrates efficient identification of effective model configurations across four case studies, reducing the manual effort required for parameter tuning and annotation.
Automating deep learning pipeline design for 3D biomedical imaging slashes manual annotation effort by intelligently leveraging predicted segmentation instances for assisted classifier training.
Deep learning-based segmentation and classification are crucial to large-scale biomedical imaging, particularly for 3D data, where manual analysis is impractical. Although many methods exist, selecting suitable models and tuning parameters remains a major bottleneck in practice. Hence, we introduce the 3D data Analysis Optimization Pipeline, a method designed to facilitate the design and parameterization of segmentation and classification using two Bayesian Optimization stages. First, the pipeline selects a segmentation model and optimizes postprocessing parameters using a domain-adapted syntactic benchmark dataset. To ensure a concise evaluation of segmentation performance, we introduce a segmentation quality metric that serves as the objective function. Second, the pipeline optimizes design choices of a classifier, such as encoder and classifier head architectures, incorporation of prior knowledge, and pretraining strategies. To reduce manual annotation effort, this stage includes an assisted class-annotation workflow that extracts predicted instances from the segmentation results and sequentially presents them to the operator, eliminating the need for manual tracking. In four case studies, the 3D data Analysis Optimization Pipeline efficiently identifies effective model and parameter configurations for individual datasets.
