Generalizable Blood Cell Detection via Unified Dataset and Faster R-CNN

TL;DR

This study develops a standardized dataset and employs Faster R-CNN with transfer learning to improve blood cell detection accuracy and stability in microscopic images, facilitating automated hematological diagnosis.

Contribution

It introduces a unified dataset from multiple sources and demonstrates the effectiveness of transfer learning with Faster R-CNN for blood cell detection.

Findings

Transfer learning significantly speeds up training convergence.

The proposed method achieves a low validation loss of 0.08666.

Unified dataset improves model robustness across heterogenous data.

Abstract

This paper presents a comprehensive methodology and comparative performance analysis for the automated classification and object detection of peripheral blood cells (PBCs) in microscopic images. Addressing the critical challenge of data scarcity and heterogeneity, robust data pipeline was first developed to standardize and merge four public datasets (PBC, BCCD, Chula, Sickle Cell) into a unified resource. Then employed a state-of-the-art Faster R-CNN object detection framework, leveraging a ResNet-50-FPN backbone. Comparative training rigorously evaluated a randomly initialized baseline model (Regimen 1) against a Transfer Learning Regimen (Regimen 2), initialized with weights pre-trained on the Microsoft COCO dataset. The results demonstrate that the Transfer Learning approach achieved significantly faster convergence and superior stability, culminating in a final validation loss of 0.08666, a substantial improvement over the baseline. This validated methodology establishes a robust foundation for building high-accuracy, deployable systems for automated hematological diagnosis.