HQ-DLHM-OD-Database

Among modern digital microscopy techniques, Digital Lensless Holographic Microscopy (DLHM) is one of the most straightforward label-free coherent imaging approaches. However, being lensless technique demands a two-step coherent imaging process to recover the object under study information; the usability of the method is commonly hindered by the need for opto-digital postprocessing of the recordings, both for retrieving the information of interest and its display adaptation to the user’s needs.
Integrating artificial intelligence (AI) and machine learning (ML) solutions into DLHM to develop automated detection systems and quality enhancement models is a promising field of application that can bridge the gap between the technique output and the user’s expectations. Nonetheless, the achievable performance of these models is directly limited by the quality and scope of the database employed in their training. Preparing a large, comprehensive, and high-quality DLHM dataset remains a difficult challenge to achieve the said solutions.

This work presents a high-quality open-source dataset of DLHM recordings and the needed information for its reconstruction. The database comprises 11720 4640x3506 holograms recorded with an 8-bit monochromatic camera supported by an MN34230 sensor with a 3.8 um pixel size. Laser diodes of 405 nm, 510 nm, and 654 nm were employed to record a diversity of biological and non-biological samples described in the dataset. A .csv file containing all the main recording parameters, the distance between the point source and the sample plane, and the illumination wavelength is added to the presented dataset folders.

在现代数字显微技术领域，无透镜全息显微术（Digital Lensless Holographic Microscopy，DLHM）是最为简便的无标记相干成像方法之一。然而，作为无透镜技术，其需通过两步相干成像流程以还原待测物体的信息；该方法的易用性通常受限于需对采集到的全息记录进行光数字后处理——这一过程既要提取目标信息，又需根据用户需求适配显示效果。将人工智能（Artificial Intelligence，AI）与机器学习（Machine Learning，ML）技术融入DLHM，开发自动化检测系统与质量增强模型，是极具应用前景的研究方向，可有效弥合该技术输出与用户预期之间的差距。然而，此类模型的可实现性能直接受限于训练所用数据库的质量与覆盖范围。要实现上述应用方案，构建大规模、全覆盖且高质量的DLHM数据集仍是一项艰巨挑战。本研究公开了一套高质量的开源DLHM采集数据集及配套的图像重建所需信息。该数据集包含11720幅分辨率为4640×3506的全息图，采集所用设备为搭载MN34230传感器的8位单色相机，传感器像素尺寸为3.8 μm。研究采用405 nm、510 nm及654 nm三种波长的激光二极管，采集了数据集描述中涵盖的各类生物与非生物样本的全息图。数据集文件夹中附带一份.csv格式文件，内含所有核心采集参数，包括点光源与样本平面的间距以及照明光波长。