Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring Few-Shot Object Detection on Blood Smear Images: A Case Study of Leukocytes and Schistocytes

Davide Antonio Mura, Michela Pinna, Lorenzo Putzu, Andrea Loddo, Alessandra Perniciano, Olga Mulas, Cecilia Di Ruberto

TL;DR

This paper tackles few-shot object detection for blood smear analysis, focusing on leukocytes and schistocytes, by evaluating the DE-ViT prototype-based open-set detector against Faster R-CNN baselines. It uses two datasets, Raabin-WBC and the newly introduced SC-IDB, and highlights domain-shift issues that undermine FSOD performance when transferring from general-object pretraining to medical images. The study provides a detailed protocol for prototype creation with a DINOv2 ViT backbone and reports that, despite some gains from domain-specific training, classical detectors remain more robust in few-shot settings for these medical tasks. The work underlines the importance of domain-adaptive data and approaches for reliable automatic blood cell enumeration and motivates future improvements in domain-focused FSOD methods.

Abstract

The detection of blood disorders often hinges upon the quantification of specific blood cell types. Variations in cell counts may indicate the presence of pathological conditions. Thus, the significance of developing precise automatic systems for blood cell enumeration is underscored. The investigation focuses on a novel approach termed DE-ViT. This methodology is employed in a Few-Shot paradigm, wherein training relies on a limited number of images. Two distinct datasets are utilised for experimental purposes: the Raabin-WBC dataset for Leukocyte detection and a local dataset for Schistocyte identification. In addition to the DE-ViT model, two baseline models, Faster R-CNN 50 and Faster R-CNN X 101, are employed, with their outcomes being compared against those of the proposed model. While DE-ViT has demonstrated state-of-the-art performance on the COCO and LVIS datasets, both baseline models surpassed its performance on the Raabin-WBC dataset. Moreover, only Faster R-CNN X 101 yielded satisfactory results on the SC-IDB. The observed disparities in performance may possibly be attributed to domain shift phenomena.

Exploring Few-Shot Object Detection on Blood Smear Images: A Case Study of Leukocytes and Schistocytes

TL;DR

This paper tackles few-shot object detection for blood smear analysis, focusing on leukocytes and schistocytes, by evaluating the DE-ViT prototype-based open-set detector against Faster R-CNN baselines. It uses two datasets, Raabin-WBC and the newly introduced SC-IDB, and highlights domain-shift issues that undermine FSOD performance when transferring from general-object pretraining to medical images. The study provides a detailed protocol for prototype creation with a DINOv2 ViT backbone and reports that, despite some gains from domain-specific training, classical detectors remain more robust in few-shot settings for these medical tasks. The work underlines the importance of domain-adaptive data and approaches for reliable automatic blood cell enumeration and motivates future improvements in domain-focused FSOD methods.

Abstract

The detection of blood disorders often hinges upon the quantification of specific blood cell types. Variations in cell counts may indicate the presence of pathological conditions. Thus, the significance of developing precise automatic systems for blood cell enumeration is underscored. The investigation focuses on a novel approach termed DE-ViT. This methodology is employed in a Few-Shot paradigm, wherein training relies on a limited number of images. Two distinct datasets are utilised for experimental purposes: the Raabin-WBC dataset for Leukocyte detection and a local dataset for Schistocyte identification. In addition to the DE-ViT model, two baseline models, Faster R-CNN 50 and Faster R-CNN X 101, are employed, with their outcomes being compared against those of the proposed model. While DE-ViT has demonstrated state-of-the-art performance on the COCO and LVIS datasets, both baseline models surpassed its performance on the Raabin-WBC dataset. Moreover, only Faster R-CNN X 101 yielded satisfactory results on the SC-IDB. The observed disparities in performance may possibly be attributed to domain shift phenomena.

Paper Structure

This paper contains 26 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. State of the Art
  3. Background
  4. Related Work
  5. Materials and Methods
  6. Dataset
  7. Raabin-WBC
  8. Preprocessing.
  9. SC-IDB
  10. Preprocessing.
  11. Evaluation Approach
  12. Prototype creation
  13. Experiments and Results
  14. Setup
  15. Comparative Models
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Image depicting the 5 types of leucocytes ref_ai2030025.
  • Figure 2: Sample images extracted from the dataset involved in this work. \ref{['fig:sub1']} presents a full-size blood smear image with some schistocytes inside the blue boxes. \ref{['fig:sub2']} is a crop of the same image presenting a schistocyte and a healthy RBC. Source: SC Ematologia e CTMO Ospedale Businco Cagliari.
  • Figure 3: Prototype creation scheme of the DE-ViT model. Source: Zhang et al. ref_devit