LUKAS HANES

May 27, 2026

I Twisted My Knee Skiing and Built MRI Scout

In April I twisted my knee skiing.

Nothing heroic. No helicopter, no dramatic mountain story. Just one bad movement, a strange feeling in the knee, and the usual human optimism: maybe it is fine.

It was not fine.

So I went through the normal process. Doctor, MRI, waiting. The MRI itself is quick. The funny part is that the scan is done immediately, but the result still takes days.

And days are plenty of time to sit at home and overthink everything.

The other funny part: they still give you the MRI data on a CD. A CD. In 2026.

I did not even have a CD drive anymore, because apparently I believed we had moved past that stage of civilization. One external drive order fixed that in a few hours.

The waiting was harder to fix. So I built a small app: MRI Scout.

MRI Scout upload and analysis interface
MRI Scout runs locally. It starts with the important bit: this is not a diagnosis. Then it lets you upload the MRI images.

The idea

The idea is simple. If you take the MRI images with you after the scan, the app can load them, pick the useful slices, and send them to an AI model for a quick orientation.

Not a diagnosis. Not a replacement for a doctor. Just a fast first look, so you are not staring at a plastic disc for several days wondering what is inside.

For my scan, MRI Scout processed 232 JPG files from the MRI export. It selected 50 representative slices and sent them through OpenAI gpt-4o. The result came back in about 90 seconds.

And it was useful.

The result

It flagged significant cruciate ligament damage, possibly even a complete rupture. It also pointed to related findings around the knee, like bone marrow signal and soft tissue edema.

MRI Scout real run summary with OpenAI gpt-4o
A real run through the same API endpoint: 232 source JPGs, 50 selected slices, and a result in about a minute and a half. The public image uses cropped scan views and removes identifying MRI export overlays.

Again, this is not where you decide treatment. That still belongs to the official report and a real doctor. But it is enough to know that this is probably not "just walk it off".

Three days later, the official result came back. MRI Scout was very close to what the report said. Not perfect, not something I would use instead of a doctor, but close enough to prove the point: waiting completely blind does not make much sense anymore.

That is the whole point.

Free if useful

MRI Scout is not trying to be your doctor. It does not wear a white coat, it does not have a waiting room, and it definitely should not decide what happens to your knee. It is just a small tool for the annoying gap between the scan and the result. The scan exists. The data exists. The AI can look at images. But the patient still waits.

If someone wants it, MRI Scout is free under the MIT license.

I do not plan to maintain it like a real product. Honestly, I hope I will not need it again. :) But if someone else does, they do not have to start from zero. Most likely, upgrading the model and maybe refreshing the UI will be enough to keep it useful even a year from now.

FoxDesk v0.3.75 Is Live

March 6, 2026

FoxDesk.org was created by Lukas Hanes as an open-source alternative to expensive SaaS platforms. Track support tickets and work logs for your team or AI agents with no per-agent fees and no vendor lock-in.

The People Who Show Up Early Rarely Stay on the Sidelines

February 27, 2026

The mood around AI right now is strange, and honestly a little funny. Not long ago, it was mostly novelty, but now every conversation swings between extreme optimism and panic. This essay explains why the real advantage still belongs to people who start building before everyone else agrees.