> For the complete documentation index, see [llms.txt](https://lwm.mit.edu/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://lwm.mit.edu/course-materials/computation-questions/module-3-pixels-to-pathway.md).

# Module 3: Pixels to Pathway

Characterizing Tumor Behavior with AI-Driven Imaging

**Date:** Friday, October 16th, 2026 (12-1:30 PM)

**Presenters**: Ashley Ruehr (HTS Core) and Alexander Greer (Vander Heiden Lab)

{% hint style="info" %}
**Central Question**: "How can analysis of imaging modalities be enhanced with computation to inform understanding of a biological endpoint?"

**Learning Objectives:**

* Understand common bottlenecks in manual imaging modalities.
* Learn how to apply machine learning tools to analyze and classify images at scale.
* Interpret neural network features to infer learned morphological properties. 
* Connect image processing capabilities to techniques for designing future experiments.
  {% endhint %}

<table data-view="cards"><thead><tr><th></th><th data-hidden data-card-cover data-type="image">Cover image</th><th data-hidden data-card-target data-type="content-ref"></th></tr></thead><tbody><tr><td><h4>Presentation Slides</h4></td><td><a href="https://2078575508-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FpK8DMWsoyo5vzRk1L6UL%2Fuploads%2Fts7OmRQRADRLvZNKaeXq%2Fm3_cover.png?alt=media&amp;token=4e42c1f0-dbcb-4184-a8cb-fee8df6d5e04">m3_cover.png</a></td><td><a href="https://docs.google.com/presentation/d/1kdruvVIb1IFHdic5etEJy5-JSZIdRGbVOEhoyBuOsPo/edit?usp=sharing">https://docs.google.com/presentation/d/1kdruvVIb1IFHdic5etEJy5-JSZIdRGbVOEhoyBuOsPo/edit?usp=sharing</a></td></tr><tr><td><h4>Code Workbooks</h4></td><td><a href="https://2078575508-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FpK8DMWsoyo5vzRk1L6UL%2Fuploads%2FFTkJDrRVoYQm9FfUne9R%2F1765109208272.png?alt=media&amp;token=27836f71-a47a-49b8-acf1-a5c2e46a8129">1765109208272.png</a></td><td><a href="https://drive.google.com/drive/folders/18yq-c-t2ccAQqbRuiit2eEcnAgyhpwwv?usp=sharing">https://drive.google.com/drive/folders/18yq-c-t2ccAQqbRuiit2eEcnAgyhpwwv?usp=sharing</a></td></tr><tr><td><h4>Reference Guide</h4></td><td><a href="https://2078575508-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FpK8DMWsoyo5vzRk1L6UL%2Fuploads%2FkqDoJFZYBOWo11m38vMr%2Fchoose-markdown.jpg?alt=media&amp;token=ac805070-e315-4d72-8f54-4bcaa5376197">choose-markdown.jpg</a></td><td><a href="https://drive.google.com/drive/folders/18yq-c-t2ccAQqbRuiit2eEcnAgyhpwwv?usp=sharing">https://drive.google.com/drive/folders/18yq-c-t2ccAQqbRuiit2eEcnAgyhpwwv?usp=sharing</a></td></tr></tbody></table>

## Demo

We explore image processing techniques using a sample imaging dataset provided by [PathMNIST](https://medmnist.com/), exploring how histological morphology reveals information about tissue samples (and when morphology might not be as important as you think!)

## Exercise

We adapt a pipeline built by the Koch Institute's High Throughput Sciences core, analyzing the cell identity of patient *ascites* samples to be used in drug response testing. Check out the [HTS Core website](https://htslab.org/) for more information on their work!

## Discussion

For any imaging model, always keep these five questions in mind:

1. **What are the "rows" and what are the "columns"**? If nobody can tell you, nobody knows what it learned.
2. **Who chose the columns**: a person, or the network?
3. **Measured on whose samples**? Same slides, or different patients, scanner, staining run, day?
4. **What does it get confidently wrong**, and when you look at those images, is it the model or the label?
5. **What would falsify it at the bench**?

Ask questions and share your own applications of these techniques in the Collaboration Park (see the **Computation Questions** channel).

{% content-ref url="/pages/nXkZZhFpJGTQPmWCg9j4" %}
[Community](/community.md)
{% endcontent-ref %}