Platelet enumeration in dense aggregates

TL;DR

This work tackles the challenge of counting platelets, especially within aggregates, by combining semantic segmentation with tailored counting methods. It explores how convolutional kernel design and explicit class separation (singular platelets vs aggregates) in U-Net architectures influence performance, and introduces the peak cluster method (PCM) for counting, benchmarked against PAM and CCA. The results show that preserving high-resolution spatial features and using a platelet-aggregate class significantly improves segmentation accuracy, and PCM provides competitive, often superior, platelet counts compared to traditional methods, particularly for larger aggregates. The findings highlight the importance of model design and specialized counting strategies for small, dense objects and suggest applicability to other domains with similar challenges.

Abstract

Identifying and counting blood components such as red blood cells, various types of white blood cells, and platelets is a critical task for healthcare practitioners. Deep learning approaches, particularly convolutional neural networks (CNNs) using supervised learning strategies, have shown considerable success for such tasks. However, CNN based architectures such as U-Net, often struggles to accurately identify platelets due to their sizes and high variability of features. To address these challenges, researchers have commonly employed strategies such as class weighted loss functions, which have demonstrated some success. However, this does not address the more significant challenge of platelet variability in size and tendency to form aggregates and associations with other blood components. In this study, we explored an alternative approach by investigating the role of convolutional kernels in mitigating these issues. We also assigned separate classes to singular platelets and platelet aggregates and performed semantic segmentation using various U-Net architectures for identifying platelets. We then evaluated and compared two common methods (pixel area method and connected component analysis) for counting platelets and proposed an alternative approach specialized for single platelets and platelet aggregates. Our experiments provided results that showed significant improvements in the identification of platelets, highlighting the importance of optimizing convolutional operations and class designations. We show that the common practice of pixel area-based counting often over estimate platelet counts, whereas the proposed method presented in this work offers significant improvements. We discuss in detail about these methods from segmentation masks.

Figures (3)

• Figure 1: Overview of segmentation and platelet counting. Images are cropped as $16 \times 256 \times 256$ images before using a U-Net network for semantic segmentation. Platelet and platelet aggregate segmentation masks are then used to count platelets using three different approaches: 1) Peaks Cluster Method (proposed method); 2) Pixel Area Method; and 3) Connect Component Analysis.
• Figure 2: High-resolution images and corresponding segmentation masks from different U-Net networks (see Table \ref{['tab:f1_scores']} for details). Platelet count for the aggregate is 14.
• Figure 3: Regression analyses of counting the PAM, PCM, and CCA methods. Regression was performed by forcing a zero y-intercept. Shaded are refers to standard deviations for respective methods.