研究目的
To benchmark registration methods on differently stained histological slides, evaluating their accuracy, robustness, and computation time.
研究成果
The study presents an experimental evaluation of publicly available image registration methods on challenging images of differently stained histological tissue. The selected registration methods cover the most common similarity criteria and optimization techniques. Although the execution time of some methods is reasonably good, the performances as measured by the proposed metrics show that the task is still not fully solved.
研究不足
The robustness does not approach 100% for any of the methods, and the mean accuracy is still far from that of a human annotator.
1：Experimental Design and Method Selection:
The study compares eleven fully automatic registration methods covering widely used similarity measures and optimization strategies with both linear and elastic transformations.
2：Sample Selection and Data Sources:
The dataset is composed of 616 image pairs of differently stained histological slides at two different scales.
3：List of Experimental Equipment and Materials:
High-resolution whole-slide images of tissue were acquired, with original sizes varying up to 45k×45k pixels.
4：Experimental Procedures and Operational Workflow:
For each method, the best parameter configuration is found and applied to all image pairs. Performance is evaluated based on registration inaccuracy, robustness, and execution time.
5：Data Analysis Methods:
The performance is evaluated from several perspectives, including registration inaccuracy on manually annotated landmarks, method robustness, and processing computation time.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
