StarDist_TumorCell_nuclei

This repository contains a StarDist deep learning model designed for segmenting tumor cell nuclei from the DAPI channel in fluorescence microscopy images while excluding HUVEC nuclei. The model was trained to accurately detect individual tumor cell nuclei for subsequent measurement of CD44, ICAM1, ICAM2, or Fibronectin intensity around or under the nuclei. The model achieved an Intersection over Union (IoU) score of 0.558 and an F1 Score of 0.793, reflecting its capability to distinguish tumor cell nuclei from HUVEC nuclei.

Specifications

Model: StarDist for segmenting tumor cell nuclei from the DAPI fluorescence channel

Training Dataset:

Number of Images: 48 paired fluorescence microscopy images and label masks

Microscope: Spinning disk confocal microscope (3i CSU-W1) with a 20x objective, NA 0.8

Data Type: Fluorescence microscopy images of the DAPI channel with manually segmented masks

File Format: TIFF (.tif)

Fluorescence Images: 16-bit

Masks: 8-bit

Image Size: 920 x 920 pixels (Pixel size: 0.6337 x 0.6337 µm²)

Model Capabilities:

Segment Tumor Cell Nuclei: Detects individual tumor cell nuclei in the DAPI channel while distinguishing them from HUVEC nuclei

Measure Intensity: Enables measurement of CD44, ICAM1, ICAM2, or Fibronectin intensity around or under tumor cell nuclei in respective channels

Performance:

Average IoU: 0.558

Average F1 Score: 0.793

Model Training: Conducted using ZeroCostDL4Mic (https://github.com/HenriquesLab/ZeroCostDL4Mic/wiki)

Reference

Fast label-free live imaging reveals key roles of flow dynamics and CD44-HA interaction in cancer cell arrest on endothelial monolayers

Gautier Follain, Sujan Ghimire, Joanna W. Pylvänäinen, Monika Vaitkevičiūtė, Diana Wurzinger, Camilo Guzmán, James RW Conway, Michal Dibus, Sanna Oikari, Kirsi Rilla, Marko Salmi, Johanna Ivaska, Guillaume Jacquemet

bioRxiv 2024.09.30.615654; doi: https://doi.org/10.1101/2024.09.30.615654