Enhanced Multi-Label Detection and Classification of Overlapping Red Blood Cells Using Modified RCNN with Transformer Augmentation

Abstract

The microscopic analysis of red blood cells (RBCs) provides critical insights into a variety of hematological conditions such as anemia, thalassemia, and sickle cell disease. Conventional manual inspection methods, while widely practiced, suffer from subjectivity, time inefficiency, and limited reproducibility. Recent advancements in artificial intelligence and deep learning have accelerated the automation of blood cell analysis; however, accurate detection and classification of overlapping RBCs remain challenging due to ambiguous boundaries and dense spatial arrangements. This research introduces a Modified Region-Based Convolutional Neural Network (RCNN) integrated with Transformer augmentation to improve multi-label detection and classification in overlapping RBC images. The proposed framework—termed MR-TNet—leverages the region proposal capabilities of RCNN alongside the global attention mechanism of Transformers to enhance contextual understanding and feature representation. Extensive experimentation on annotated blood smear datasets demonstrates the superior performance of MR-TNet over traditional CNN and RCNN architectures in metrics such as accuracy, precision, recall, and F1-score. The model also maintains computational efficiency suitable for real-world deployment in pathology workflows. Beyond its diagnostic utility, the proposed framework lays the groundwork for integrating automated analysis of multiple hematological components, enabling scalable, AI-driven clinical decision support systems.