POSEA: A novel algorithm to evaluate the performance of multi-object instance image segmentation.

Area of Science:

• * Microscopy image analysis
• * Computational biology
• * Image segmentation evaluation

Background:

• * Existing cell segmentation methods lack robust evaluation for complex images.
• * Pixel-level comparison is insufficient, failing to detect errors between adjacent segmented objects.
• * Accurate segmentation evaluation is critical for quantitative biological studies.

Purpose of the Study:

• * To introduce a novel algorithm, POSEA, for accurate per-object segmentation evaluation.
• * To compare POSEA's performance against traditional pixel-level evaluation methods.
• * To demonstrate POSEA's utility in analyzing microscopy images of biological samples.

Main Methods:

• * Development of a per-object segmentation evaluation algorithm (POSEA).
• * Calculation of precision, recall, and f-measure metrics for each segmented object.
• * Validation using simulated and real fluorescence microscopy images of diverse cell types.

Main Results:

• * POSEA yields lower accuracy metrics than pixel-level evaluation due to precise error accounting.
• * The algorithm correctly identifies and penalizes misclassified pixels shared between adjacent objects.
• * Demonstrated robustness across simulated and multiple biological sample datasets.

Conclusions:

• * POSEA provides more accurate segmentation evaluation metrics than standard pixel-level approaches.
• * The algorithm is essential for applications requiring precise segmentation of unique adjacent objects.
• * POSEA enhances the reliability of quantitative analysis in cell biology research.