Bayesian Optimization for Design Parameters of 3D Image Data Analysis

TL;DR

This paper presents a Bayesian Optimization pipeline that automates the selection and tuning of models and parameters for 3D biomedical image segmentation and classification, reducing manual effort and improving performance.

Contribution

The introduced pipeline combines two Bayesian Optimization stages with a novel segmentation quality metric and an assisted annotation workflow for efficient 3D image data analysis.

Findings

Efficient identification of optimal models and parameters across four case studies.

Reduction in manual annotation effort through automated instance extraction.

Improved segmentation and classification performance on biomedical 3D datasets.

Abstract

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.