AI-assisted holographic fluorescence protocol for discriminating and counting microplastics and microalgae in real samples

Microplastic contamination in aquatic ecosystems poses a specific analytical problem: in real matrices, plastic particles coexist with cells and organic matter of similar morphology and optical properties, so their reliable discrimination and counting remain the main bottleneck, aggravated by the reliance on manual counting. This work presents an AI-assisted analytical protocol, implemented in HoloDetect, for the detection, discrimination and quantification of microplastics and microalgae in aquatic samples, based on the integration of digital holographic microscopy with multi-wavelength fluorescence imaging. For each field of view, holographic and fluorescence images of the same measurement volume are acquired and spatially registered, so that every particle is characterised by combining both sources. The analytical advantage lies in this combination: holographic descriptors provide geometry, depth and focus, while fluorescence separates photosynthetic cells from plastic particles. The fluorescence module incorporates excitation lasers at 405, 450 and 532 nm and an emission filter with a 550 nm cut-off wavelength. Object-level descriptors included area, equivalent diameter, length, width, perimeter, circularity, elongation, aspect ratio, depth, focus-related parameters, and fluorescence intensity and area across multiple spectral channels. The workflow comprises image acquisition, object segmentation, feature extraction, supervised classification and quantitative object-level analysis. As a supporting tool rather than the main methodological novelty, a classification module is trained on manually selected representative images assigned to predefined classes, adaptable to the target categories of each sample: the system returns class assignments, confidence values and object-level measurements, and its performance was assessed using accuracy, precision, recall, F1 score and confusion matrices. Validation was carried out at two levels. In synthetic suspensions of known composition containing polyethylene (PE) microplastics, Chlorella sorokiniana and Scenedesmus almeriensis at 1:1:1, 1:2:1, 1:1:2 and 2:1:1 ratios, the protocol correctly detected proportional changes in particle abundance when individual microalgal concentrations were increased. In an environmental sample from an open-channel system, spiked with PE microplastics and analysed using Nile Red fluorescence to improve discrimination from microalgal cells, 449 PE particles were identified with 99.1 % precision and 92.9 % recall. The protocol reduces manual counting and delivers quantification at the individual-particle level, applicable to environmental monitoring, microplastic tracking, microalgal cultivation assessment and wastewater analysis. The authors acknowledge the financial support of the Human Resources Program of the Brazilian National Agency for Petroleum, Natural Gas and Biofuels (PRH-ANP/PRH36-UFBA) and of the REMAREC project funded by CNPq (process no. 420945/2025-3).