MaxViT-Backbone Mask R-CNN for Bacterial Cell Detection and Instance Segmentation: Toward Antimicrobial Resistance Analysis

Antimicrobial resistance (AMR) is one of the most significant and pressing threats to public health today, causing serious health issues and hindering the effective prevention and treatment of infectious diseases. Traditional culture-based methods possess several limitations, including long turnaround times and potential discrepancies in interpretation between standards. Deep learning-based segmentation and detection techniques can facilitate the analysis of bacterial cell morphology, spatial distribution, clustering, and aggregation patterns, which are important for AMR research. In this work, we propose a hybrid end-to-end instance segmentation framework that performs bacterial cell detection, localization, and instance segmentation within a single architecture, using the Multi-Axis Vision Transformer (MaxViT) as the backbone for the Mask R-CNN model. Unlike traditional convolutional neural network (CNN)-based approaches, our proposed framework captures global contextual information while preserving fine-grained morphological patterns in microscopic imagery. The framework was evaluated on a clinical microbiological rapid on-site evaluation (M-ROSE) dataset comprising four bacterial classes: Gram-negative cocci, Gram-positive cocci, Gram-negative bacilli, and Gram-positive bacilli. For segmentation, it achieved an IoU of 0.6443, a Dice score of 0.7179, and an mAP50 of 0.8225. For object detection, it achieved a precision of 0.7995, a recall of 0.8171, an F1-score of 0.8054, and an mAP50 of 0.8235. To improve visual interpretability, we also implement ROI-based Class Activation Mapping (ROI-CAM), which highlights the areas most influential to the model’s predictions. Although the proposed instance segmentation framework does not directly infer AMR, it provides a foundational component for AI-driven bacterial cell analysis within a broader AMR screening and decision-support pipeline. This framework could facilitate subsequent phenotypic characterization and AMR prediction in future studies.

Authors

Institutions

Publication Details

Journal
Electronics
Published
2026-11-17
DOI
https://doi.org/10.3390/electronics15225349
Primary Topic
Cell Image Analysis Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

MaxViT-Backbone Mask R-CNN for Bacterial Cell Detection and Instance Segmentation: Toward Antimicrobial Resistance Analysis

Md. Zasim Uddin, Khondaker Miraz Rahman, Md Atiqur Rahman Ahad, Ali Akbar et al.
3 links
Electronics
Cell Image Analysis Techniques
article

MaxViT-Backbone Mask R-CNN for Bacterial Cell Detection and Instance Segmentation: Toward Antimicrobial Resistance Analysis

Md. Zasim Uddin, Khondaker Miraz Rahman, Md Atiqur Rahman Ahad, Ali Akbar, Md. Intikhab Shahriar Hasan
article en

Abstract

Antimicrobial resistance (AMR) is one of the most significant and pressing threats to public health today, causing serious health issues and hindering the effective prevention and treatment of infectious diseases. Traditional culture-based methods possess several limitations, including long turnaround times and potential discrepancies in interpretation between standards. Deep learning-based segmentation and detection techniques can facilitate the analysis of bacterial cell morphology, spatial distribution, clustering, and aggregation patterns, which are important for AMR research. In this work, we propose a hybrid end-to-end instance segmentation framework that performs bacterial cell detection, localization, and instance segmentation within a single architecture, using the Multi-Axis Vision Transformer (MaxViT) as the backbone for the Mask R-CNN model. Unlike traditional convolutional neural network (CNN)-based approaches, our proposed framework captures global contextual information while preserving fine-grained morphological patterns in microscopic imagery. The framework was evaluated on a clinical microbiological rapid on-site evaluation (M-ROSE) dataset comprising four bacterial classes: Gram-negative cocci, Gram-positive cocci, Gram-negative bacilli, and Gram-positive bacilli. For segmentation, it achieved an IoU of 0.6443, a Dice score of 0.7179, and an mAP50 of 0.8225. For object detection, it achieved a precision of 0.7995, a recall of 0.8171, an F1-score of 0.8054, and an mAP50 of 0.8235. To improve visual interpretability, we also implement ROI-based Class Activation Mapping (ROI-CAM), which highlights the areas most influential to the model’s predictions. Although the proposed instance segmentation framework does not directly infer AMR, it provides a foundational component for AI-driven bacterial cell analysis within a broader AMR screening and decision-support pipeline. This framework could facilitate subsequent phenotypic characterization and AMR prediction in future studies.

ElectronicsVol. 15(22)
University of East London (GB), Begum Rokeya University (BD), King's College London (GB), University of Dhaka (BD)
Good health and well-being
Cell Image Analysis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon