Deeptosis: Help

✦ 1. What is Deeptosis?

Deeptosis is a web-based tool for automated, label-free identification of cell death modalities from bright-field microscopy images. It segments cells, classifies each single cell as apoptosis, pyroptosis, or other, and returns both visual and tabular results.

✦ 2. What data can I upload?

Image type: bright-field (label-free) microscopy images

Formats: PNG, JPG/JPEG, TIFF, BMP

Content: one image may contain multiple cells

Quality: in-focus images with reasonable contrast work best

Not recommended: heavily out-of-focus images, large empty areas, images from devices very different from the training domain

✦ 3. How to use the website

3.1 Upload

Click Upload and select one or more bright-field images from your local computer.

3.2 Segmentation (automatic)

The system first runs Cellpose to segment the whole image and extract single-cell crops. This step is done automatically on the server and corresponds to the Figure 1.

3.3 Classification

Each cropped cell is passed to our fine-tuned Vision Transformer model. The model predicts one of three categories: Apoptosis Pyroptosis Other (live cells, debris, poorly segmented cells, or morphologically ambiguous cases)

The model architecture include patch embedding, transformer encoder blocks, and the classification head, is shown in Figure 2A

3.4 View and download results

When processing finishes, the website displays: 1.Annotated image – original image overlaid with color-coded boxes/masks oGreen: apoptosis oRed: pyroptosis 2.Summary panel – total cells detected, per-class counts and percentages, average confidence 3.CSV file – one row per cell: filename, cell ID, bounding-box coordinates, predicted class, class name, confidence score You can download the CSV for downstream analysis or to match with experimental metadata.

3. 5 – Web interface and job view

The web page is organized into an input/console area and an output/result area. Input panel (Figure 3A): users can drag and drop standard images or click the Upload Image button. The Parameter block lets you show both classes or focus on apoptosis/pyroptosis only. The Threshold slider (default 0.50) controls the minimum confidence for displaying a prediction. After configuration, click Submit to start the backend pipeline (Cellpose segmentation → Deeptosis classification). Result panel (Figure 3B): once the job is finished, the interface shows a summary table listing the image file, total detected cells, counts and percentages for each class, mean confidence, and confidence range. A Download button provides the annotated images and CSV summaries. Below the table, the Raw Image and Annotated Image are displayed side by side so that users can visually verify the predictions. Job identifiers and real-time status help track multiple submissions.

✦ 4. Performance

The current deployed model was trained on 26,000+ manually labeled single-cell images and evaluated with 5-fold cross-validation. It achieves AUROC > 0.98 for all three classes and shows very low confusion between apoptosis and pyroptosis. Quantitative metrics (ROC curves and confusion matrix) are shown in Figure 2B–2D.Note: performance may drop on images that are far from the training domain (different microscope, different illumination, very dense cultures).

✦ 5. Frequently Asked Questions (FAQ)

Q1. The system did not detect cells. What should I do?

Check image focus and brightness. Very dark, very bright, or very small cells may not be segmented. Try re-acquiring the image at a slightly higher resolution.

Q2. Why are some cells labeled as “other”?

“Other” is a mixed category for non-classical cases: live cells, debris, segmentation errors, or cells that do not clearly match the apoptosis/pyroptosis morphology learned by the model.

Q3. Can the tool identify ferroptosis or necroptosis? Not in this version. The current model exposes three outputs only: apoptosis, pyroptosis, other.

Q4. Can I upload many images at once?

Yes. Each image will be processed and shown with its own visualization and CSV output.

Q5. Can I use these predictions directly as biological proof?

Use the results as high-throughput morphological screening. For definitive conclusions, please confirm with wet-lab assays (e.g. caspase activity, gasdermin cleavage, LDH release).

✦ 6. Contact / Feedback

If you encounter wrong segmentation, obvious misclassification, or would like to contribute new cell-death categories (e.g. ferroptosis), please send the original image together with the output CSV to the development team so that the model can be improved in future updates. For technical issues and data contribution, please contact hwang@hnu.edu.cn, xuhaodong@csu.edu.cn, and wangliming@hnu.edu.cn.

Deeptosis

Deep Learning-Based Platform for Label-Free Discrimination of Apoptosis and Pyroptosis

Version 1.0

1. Workfolw »

2. Usage »

3. Information »

4. References »

5. Contact »