🎓Medical images are machine outputs, not holiday photos.
In our computer vision pipelines, we standardize to 224×224 bilinear—no crop—at both train & predict.
mlhs.ink/Playground/#SIC mlhs.ink/Playground/#SVC
Identical steps, no drift, reproducible results. #MedAI #DigitalHealth

🧩 In medical ML models, as in causal inference, not all variables are equal:

Mediators transmit the effect.

Confounders bias the effect.


Mediators should be excluded, confounders should be utilized.

👉 Explore your health data: build no-code Web ML models:
🔗 mlhs.ink/Playground/#SDC

#MedAI #CausalInference #DigitalHealth #WebML

🧩The real cause of morbidity is often the interaction between exposures — whereas predictors alone are not necessarily harmful.

We know:
1️⃣ Cortisol/adrenaline stress → reduced life expectancy
2️⃣ Skydiving → cortisol/adrenaline surge
3️⃣ But… Skydiving ≠/? Life expectancy directly

👉 Only the combination of cortisol/adrenaline and exposure time drives the outcome — morbidity.

📖 For more on interactions, read this great paper by Tyler J. VanderWeele: Interaction Tutorial (PDF)

#CausalInference #MedAI #DigitalHealth

☸️𝗦𝗔𝗜𝗠𝗦𝗔𝗥𝗔 𝗽𝗮𝗽𝗲𝗿𝘀 → 𝗻𝗼𝘄 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱𝗮𝗯𝗹𝗲 𝗮𝘀 𝗣𝗗𝗙.

No more “only URL/TXT”. Just a proper paper you can read & share.

saimsara.com

#MedAI #EvidenceBasedMedicine