A Digital Holography-Based Dataset of Red Blood Cells for Machine and Deep Learning Analysis of Hereditary Anemias
- Marika Valentino
- Cosimo Ieracitano
- Nadia Mammone
- Maria Pia Pierro
- Zhe Wang
- Anthony Iscaro
- Antonella Nostroso
- Immacolata Andolfo
- Roberta Russo
- Vittorio Bianco
- Pasquale Memmolo
- Carlo Morabito
- Pietro Ferraro
- Lisa Miccio
2026-08-18
Hereditary Anemias (HA) are genetic blood disorders affecting a large number of subjects and can cause important clinical complications. Correct, rapid and automatic identification of each type of anemia is pivotal for proper treatment and patient management. Digital holographic microscopy provides label-free quantitative phase measurements of red blood cells (RBCs) morphology. However, standardized and reproducible computational workflows for feature extraction and RBCs classification are still scarce. Here we present a collection of roughly 4600 holographic phase-contrast maps (PCMs) representing RBCs from healthy controls and patients affected by five different HA subtypes. Furthermore, a complete and openly available set of MATLAB scripts is released to implement an analysis workflow, taking PCMs as input. The workflow includes handcrafted feature extraction, as well as different conventional machine and deep learning models. Notably, deep learning architectures are trained directly on PCMs. By releasing the full pipeline, from raw phase maps to trained models, this work delivers an open and extensible framework that facilitates method comparison, ensures reproducibility, and supports the development of new approaches.