For Blueprint Reviewers, Final Version is what our end CAD model is. Design is goose.f3d present in the same file. BOM is present in root dir, table format (collapsed) at bottom of readme and PCB folder. PCB Folder contains gerbers.zip and design files. Firmware is in firmware and software in software. Everything else is in ML or doc. Check Sections for descriptions. thank you for reviewing our project :)

This is the main repo for the Arryhtmix Project, which aims to develop a novel, end-to-end framework for the prediction of cardiac eventss

We have developed a novel end to end pipeline for the prediction of cardiac arrests through machine learning alongside a medical device to facilitate the collection of data for the model.

The initial process takes place on the device. The single-lead AD8232 Sensor with integrated hardware high pass and low pass filters collects ECG Data from one lead with one electrode for common mode rejection. The data is read at a rate of 360 Hz by the nRF52840 using a 12-bit ADC and concurrently denoised through an integrated LSM6DS3. While the device is idle, it is kept in deep sleep mode, significantly reducing power consumption. Upon the collection of 10 data points, the buffer is dispatched using BLE to the host device..

Upon being dispatched to the host device, two models are run on it. First, an Arrhythmia classification model based on the MIT-BIH dataset and architected with a BiLSTM and Random Forest classifier returns a dataclass which describes the type of arrhythmia if any. Secondly, a novel hybrid CNN-LSTM model trained on the Sudden Cardiac Arrest Holter Database predicts ventricular fibrillation risk across four temporal categories. The models have been quantised for edge deployment and offer platform flexibility. A f-score of 99 percent and of 85 percent is obtained from both models respectively.

• ISEF Abstract

• Scripts
  • Miscellaneous
    • mixture of normal and vibecoded scripts made for testing
  • Rerun
    • Scripts and Blueprints for the Rerun visualizer
• Hardware
  • PCB Documents
  • CAD Files
    • all of these are from some iteration or the other, but they should represent the totality of our cad.
    • Final Version is what our end model is. Design is goose.f3d present in the same file
• Firmware
  • firmware for the device
• ML
  • Our own model training stuff, and tweaked versions of other's
• Documentation
  • Extra stuff relating to the project. Will add slide decks soon:tm:

Machine Learning

Confusion Matrix

train and valuation graphs

Software

Machine