Temporal dynamics improves machine learning-based prediction of cell state from quantitative phase imaging

In plain English

The model watched cells change over time instead of judging each cell from one still image.

What they found

• The largest accuracy gain came between one and three frames.
• The three-dimensional model continued improving through eleven frames and reached 96.5% accuracy.
• Fast events such as mitosis needed less temporal context than ferroptosis.

The catch

• The record is a preprint and has not completed peer review.
• This summary is based on the abstract; methods and supplementary analyses were not independently rechecked.

Methods at a glance

Study type

Computational image-classification study

Biological system

Six live-cell states captured by quantitative phase imaging

Organism

Cell model not specified in the abstract

Dataset

Quantitative phase imaging timelapse sequences

Sample size

1,874 timelapse sequences

Model architecture

Two-dimensional CNNs and a three-dimensional spatiotemporal CNN

Training objective

Classify interphase, mitosis, arrest, apoptosis, ferroptosis, and necroptosis

Code

Not reported in the abstract

Data

Not reported in the abstract

Tasks and metrics

Tasks: Six-class cell-state classification

Metrics: Accuracy; Error rate

Baselines: Single-frame 2D CNN; Multi-frame 2D CNN

Principal findings

• The 3D model reached 96.5% accuracy and a 3.5% error rate with eleven frames.

What remains unresolved

External validation: Not reported

• The record is a preprint and has not completed peer review.
• This summary is based on the abstract; methods and supplementary analyses were not independently rechecked.