We break down a neural network into a sequence of modules...
🌳 -->🍃🍃🍃🍃🍃🍃🍃🍃🍃🍃🍃🍃🍃
...to create MoDN: Modular Clinical Decision Support Networks.
👆 Check out our publication in PLOS Digital Health 📜 to see how MoDN allows the flexibility to adapt to volatile resource availability in constrained settings, where the clinician:
1) Could compose models at the bedside using whatever number or combination of inputs (questions/tests) are available to them #composability #mixnmatch 🔀
2) Add new inputs without needing to retrain the entire model #compartmentalization 🎡
3) Learn from the notoriously flawed data derived from clinical decision support tools #systematicmissingness 👻
🏆 Phenomenal work by studentpower: Cécile Trottet and Thijs Vogels
🏥 Thank you to our clinical collaborators Alexandra Kulinkina, Rainer Tan, Ludovico Cobuccio
--
❤️ This work was funded by Fondation Botnar as part of the Dynamic Study