AneRBC dataset: a benchmark dataset for computer-aided anemia diagnosis using RBC images.

Area of Science:

• Medical Image Analysis
• Hematology
• Artificial Intelligence

Background:

• Diagnosing anemia relies on visual analysis of red blood cell (RBC) morphology in peripheral blood smears.
• The lack of comprehensive anemic RBC datasets impedes the development of deep convolutional neural networks (CNNs) for automated RBC morphology analysis.

Purpose of the Study:

• To introduce a benchmark dataset, Anemic RBC (AneRBC), to facilitate the training and validation of CNN models for RBC morphology analysis in anemia.

Main Methods:

• The AneRBC dataset was created in two versions: AneRBC-I (1000 high-resolution images with ground truth) and AneRBC-II (12,000 sub-images).
• Images include healthy and anemic RBCs with detailed annotations, complete blood count, and morphology reports.
• Four state-of-the-art CNN models were employed for segmentation and classification to validate the dataset's utility.

Main Results:

• AneRBC-I contains 500 healthy and 500 anemic images, detailing approximately 1,550,000 RBC elements.
• AneRBC-II comprises 12,000 images, derived from AneRBC-I by dividing high-resolution images into sub-images.
• The dataset facilitates validation of CNN model performance against clinical data.

Conclusions:

• The AneRBC dataset addresses the critical need for standardized data in developing AI for anemia diagnosis.
• This resource supports advancements in computer-aided diagnosis of RBC morphological deformities.
• The dataset enables robust training and testing of CNNs for improved anemia detection.