MSc Thesis Defense: Miray Esra Palaz, U–NET–BASED QUANTIFICATION OF MORPHOLOGICAL ADAPTATIONS IN CIPROFLOXACIN–EXPOSED ESCHERICHIA COLI CELLS IN COMPLEX MICROENVIRONMENTS, Date & Time: 07 July, 2026 – 5:00 PM, Place: FENS L027
U–NET–BASED QUANTIFICATION OF MORPHOLOGICAL
ADAPTATIONS IN CIPROFLOXACIN–EXPOSED ESCHERICHIA
COLI CELLS IN COMPLEX MICROENVIRONMENTS
Miray Esra Palaz
Molecular Biology, Genetics and Bioengineering, MSc Thesis, 2026
Thesis Jury
Assoc. Prof. Meltem Elitaş Hartmann (Thesis Advisor)
Assoc. Prof. Özlem Kutlu
Assoc. Prof. Aşkın Kocabaş
Date & Time: 7th July, 2026 – 5:00 PM
Place: FENS L027
Zoom: https://sabanciuniv.zoom.us/j/
Keywords : U-Net, cell segmentation, Escherichia coli, ciprofloxacin, antimicrobial resistance
Abstract
Antimicrobial resistance is a pressing global public health challenge, necessitating rapid, reliable antibiotic susceptibility testing. Conventional antibiotic susceptibility testing methods rely on population-level measurements and often require prolonged incubation times, limiting their ability to capture single-cell-level phenotypic heterogeneity. Phase-contrast microscopy enables label-free observation of antibiotic-induced morphological changes in individual bacterial cells, offering an alternative for rapid phenotypic testing. However, accurate, automated segmentation of bacterial cells remains challenging under heterogeneous imaging conditions that closely resemble clinical samples. This thesis presents CiproSeg, a U-Net-based deep learning framework for robust segmentation of Escherichia coli cells in phase-contrast microscopy images and quantitative analysis of ciprofloxacin-induced morphological responses at the single-cell level. The model was developed using a manually annotated dataset representing diverse imaging conditions and evaluated for its ability to generalize across heterogeneous experimental settings.Experimental results demonstrate that CiproSeg achieved a Dice coefficient of 0.768, outperforming the widely used Cellpose and MiSiC segmentation frameworks under heterogeneous imaging conditions. Morphological analysis showed ciprofloxacin treatment induced significant bacterial cell elongation, consistent with the inhibition of cell division through activation of the bacterial stress response. Contrarily, this morphological response could not be reliably detected under heterogeneous imaging conditions, highlighting the limitations of existing segmentation methods for clinical applications.These findings demonstrate that CiproSeg enables accurate and robust bacterial cell segmentation across diverse imaging conditions and facilitates reliable single-cell morphological analysis for label-free antibiotic susceptibility testing. Proposed framework represents a step toward automated, rapid, clinically applicable phenotypic antibiotic susceptibility testing based on microscopy and deep learning.