Before I get into this week’s issue, one of my peers is running a survey and needs people to fill it out! It only takes 5-7 minutes. See info below:

“I’m running a national survey on technology adoption, virtual care, and AI in PT and OT practice. The goal is to build a real, data-driven benchmark for rehabilitation professionals.

Everyone who participates gets the executive summary before it goes public.

We’re also doing a drawing for participants:

Take the Survey

I've spent the last several issues being critical. The data center costs, the water, the bills, the forced adoption, the loss-of-control incident, the companies lobbying to dodge accountability. All of that criticism stands. I'm not walking any of it back.

And yet. This is the same technology that is quietly transforming cancer research and detection in ways that are genuinely saving lives right now. Precision medicine. Predictive analytics. Not in a demo. Not in a pitch deck. In real hospitals, on real patients.

Holding both of these at once is not a contradiction. It's reality. The version of AI being force-fed to us through every consumer app, trained on scraped data and cooled by strained water supplies, is worth criticizing hard. The version being used to map tumors at single-cell resolution and catch cancers a human eye would miss is worth celebrating. Refusing to acknowledge the second because you're angry about the first isn't principled. It's just incomplete.

This week's headline is the celebrating kind.

Researchers at the University of Southampton, working with University Hospital Southampton, built an open-source AI tool called CenSegNet that can analyze hundreds of thousands of cells inside tumor samples with a speed and precision no human could match. They used it to study breast cancer tissue and uncovered patterns that were, until now, essentially invisible.

Inside your cells are tiny structures called centrosomes. Think of them as the cell's organizing hubs. They help cells divide correctly. When they go haywire, cells start accumulating genetic errors, which is a hallmark of cancer. Scientists have known this for over a century, but actually studying these defects across a whole tumor, cell by cell, is too complex to do by hand. CenSegNet does it at single-cell resolution across an entire tumor sample.

Analyzing more than 330,000 centrosomes across 911 tumor specimens from 127 patients, the AI found something humans had missed: what everyone assumed was one single type of abnormality is actually two distinct kinds that behave independently and even occupy different regions of the same tumor. Tumors with more of the enlarged-centrosome type were linked to more aggressive disease, higher grade, and lymph node involvement. Patients with fewer of them in the tumor core tended to have better survival.

That's not a small finding. It points toward new biomarkers, better risk prediction, and eventually, more personalized treatment matching. Isn’t that what it’s all about??

Watch the gap between "research breakthrough" and "bedside tool," because it's usually years, not months. The honest framing here is promise, not arrival. Watch also for how open-source science accelerates this. Because the tool is free, the pace of validation across other labs and other cancers could move faster than a proprietary tool ever would.

TL;DR: We know and believe that early detection = early intervention = better outcomes = better quality of life. You add in precision medicine, predictive analytics, and personalized treatment. That’s what this is about.

The longer version:
At first glance, tumor cell analysis feels a world away from occupational therapy. But follow the timeline. Better risk prediction and earlier, more precise diagnosis change when and how patients enter the care continuum, and OT lives all along that continuum. Earlier detection of aggressive disease can mean earlier intervention, which shifts rehab timelines, discharge planning, and caregiver readiness. More personalized treatment matching can mean different side effect profiles, which changes what we're helping patients adapt to and recover from.

There's also a bigger-picture point that ties to the hot take. This is what AI aimed at the right problem, by the right people, looks like. It's not replacing the pathologist or the oncologist. It's giving them sight they didn't have before. That's the model of AI integration OT should be advocating for across healthcare: not automation that removes the human, but augmentation that sharpens the human's judgment. When we argue for how AI should show up in our own practice, this is the template worth pointing to.