Towards Cross-Domain Single Blood Cell Image Classification via Large-Scale LoRA-based Segment Anything Model

TL;DR

This work tackles cross-domain blood cell image classification by integrating a LoRA-finetuned Segment Anything Model (SAM) with a cross-domain autoencoder to learn intrinsic image features while suppressing domain-specific artifacts. It yields embeddings $e_i$ from LoRA-SAM and 50-dimensional latent representations $z_i$ via the autoencoder, optimized with $L= L_{ssim}+ \\lambda L_{mmd}$ and $\\\\\\lambda=5$, and trains RF, SVM, ANN, and XGBoost classifiers on $z_i$. The approach achieves state-of-the-art cross-domain results on Matek-19 and Acevedo-20, significantly outperforming baselines, and demonstrates robust generalization across domain shifts in hematology imaging. The method is practical for clinical workflows, with publicly available source code and substantial improvements in cross-domain reliability for blood cell analysis.

Abstract

Accurate classification of blood cells plays a vital role in hematological analysis as it aids physicians in diagnosing various medical conditions. In this study, we present a novel approach for classifying blood cell images known as BC-SAM. BC-SAM leverages the large-scale foundation model of Segment Anything Model (SAM) and incorporates a fine-tuning technique using LoRA, allowing it to extract general image embeddings from blood cell images. To enhance the applicability of BC-SAM across different blood cell image datasets, we introduce an unsupervised cross-domain autoencoder that focuses on learning intrinsic features while suppressing artifacts in the images. To assess the performance of BC-SAM, we employ four widely used machine learning classifiers (Random Forest, Support Vector Machine, Artificial Neural Network, and XGBoost) to construct blood cell classification models and compare them against existing state-of-the-art methods. Experimental results conducted on two publicly available blood cell datasets (Matek-19 and Acevedo-20) demonstrate that our proposed BC-SAM achieves a new state-of-the-art result, surpassing the baseline methods with a significant improvement. The source code of this paper is available at https://github.com/AnoK3111/BC-SAM.

Figures (3)

• Figure 1: The pipeline of blood cell classification. It mainly consists of image representation learning with a segment anything model (SAM) and classification model building with four widely used machine learning methods.
• Figure 2: The category distribution of blood cell image datasets.
• Figure 3: The architecture of BC-SAM. It consists of two components: LoRA-based segment anything model and cross-domain autoencoder. Snowflake refers to the frozen parameters while flame refers to the learnable parameters.