[Paper Review] Multi-Temporal Frames Projection for Dynamic Processes Fusion in Fluorescence Microscopy

Hassan Eshkiki, Sarah Costa+4 more|arXiv (Cornell University)|Jan 15, 2026

Cell Image Analysis Techniques0 citations

TL;DR

This paper presents a modular framework that fuses multi-temporal fluorescence microscopy frames into high-quality 2D images using preprocessing and z-projections, evaluated on cardiac cell datasets with 111 configurations and achieving a 44% average increase in cell count over prior methods.

ABSTRACT

Fluorescence microscopy is widely employed for the analysis of living biological samples; however, the utility of the resulting recordings is frequently constrained by noise, temporal variability, and inconsistent visualisation of signals that oscillate over time. We present a unique computational framework that integrates information from multiple time-resolved frames into a single high-quality image, while preserving the underlying biological content of the original video. We evaluate the proposed method through an extensive number of configurations (n = 111) and on a challenging dataset comprising dynamic, heterogeneous, and morphologically complex 2D monolayers of cardiac cells. Results show that our framework, which consists of a combination of explainable techniques from different computer vision application fields, is capable of generating composite images that preserve and enhance the quality and information of individual microscopy frames, yielding 44% average increase in cell count compared to previous methods. The proposed pipeline is applicable to other imaging domains that require the fusion of multi-temporal image stacks into high-quality 2D images, thereby facilitating annotation and downstream segmentation.

Motivation & Objective

• Address noise, temporal variability, and inconsistent visibility in fluorescence microscopy recordings.
• Develop a preprocessing-plus-projection pipeline that preserves biological content while enabling downstream segmentation.
• Create a modular framework that can be extended with additional preprocessing and projection techniques.
• Demonstrate the approach on dynamic cardiac cell networks with oscillatory Ca2+ signals to produce informative 2D representations.

Proposed method

• Isolate frames from video to form multi-temporal stacks and apply a three-step preprocessing pipeline (equalisation, pixel intensity remapping, noise filtering).
• Fuse preprocessed frame stacks using six z-projection techniques (MIP, AP, SP, PDP, SDP, QP).
• Evaluate a broad configuration set (111 preprocessing variants × 6 projections = 666 pipelines) to identify effective combinations.
• Employ no-reference image quality assessment (NR-IQA) metrics (NIQE, PIQE, BRISQUE) to select high-quality projections in the absence of ground-truth images.
• Provide a modular framework that can incorporate additional preprocessing methods and projection operators.
• Report results on a dataset of 91 CLSM HL-1 cardiac cell network videos with 150 frames per video.

Figure 1 : Systematic pipeline for multi-temporal frames fusion in (a) live-sample FM datasets via combined (b) image processing and (c) z-projection methods.

Experimental results

Research questions

• RQ1 Can multi-temporal frame fusion via projection techniques improve image quality and retain temporal information in fluorescence microscopy?
• RQ2 Which combinations of preprocessing steps and z-projections yield the best qualitative and NR-IQA-based quality?
• RQ3 Do projected representations facilitate downstream annotation and segmentation better than traditional frame-based approaches?
• RQ4 Is the approach generalizable to other imaging domains requiring multi-temporal fusion?
• RQ5 How do NR-IQA metrics correlate with expert perceptual quality in microscopy images?

Key findings

• The framework produced a 60,606 pipeline results across 91 videos, identifying SP and AP projections as offering strong NR-IQA performance for PIQE and BRISQUE.
• MIP generally yielded lower quality scores across NR-IQA metrics compared with SP, AP, QP, and SDP.
• PDP projections yielded unusable results due to artefacts and were discarded from further analysis.
• The approach achieved a 44% average increase in cell count compared with previous methods on the HL-1 cardiac cell dataset.
• The NR-IQA metrics showed varying sensitivity to preprocessing–projection combinations, with NIQE and BRISQUE correlating more with artefacts introduced by certain pipelines, and PIQE capturing blur and detail loss.
• The authors emphasize a fully modular and explainable pipeline aimed at enabling biologists to inspect and annotate outputs and to feed downstream DL-based segmentation.

Figure 2 : Box plot of NR-IQR scores by projection method. Yellow-filled diamond-shaped dots within the boxes indicate mean values for each projection group.