Inspiration

People that suffer from chronic back pain will go to a doctor or physiotherapist for solutions. Common recommendations are massage therapy, exercise and if the pain persists, painkillers. Healthcare professionals try to identify the source of pain by asking questions about the patient’s lifestyle and observing their body structure and posture. However, physiotherapists and doctors can only see you while you are in the clinic, and lose track of you for the rest of the day. A personalized posture-tracking T-shirt follows the patient during the day. The data helps the physiotherapist identify bad posture habits that can be corrected. Machine learning allows the construction of predictive algorithms that give decision support to the healthcare professional.

What it does

We tackle 3 sides of the problem: consumer facing, which is the patient, manufacturing, which is researchers and mass producable tshirts, and diagnosis, which is helping doctors provide rehabilitation for patients.

  1. Patient Our product interfaces with our smartphone via blutooth that is capable of recording and transmitting information from strategically placed sensors on the patients back to not only your smartphone for patients viewing and to adjust thresholds according the patients age, height, size and gender for personalized treatment.

  2. Mass producable We have created a circuit that uses a microcontroller that reads data from accelerometers and flex sensors, and processes them based on instructions transmitted via Bluetooth module. We then run a script in MatLab that gathers and processes the data from the ports. It produces recognizable physical values from raw numbers.

We then classify them according to 11 base postures based on ones found on the NHS website to treat back illnesses and built a classifier based on those data sets.

  1. Doctors We then used the python library MatPlotLib to plot a triangle that represents the patients back over a standardizes back vector over time. For example, if a patient wore this shirt for 1 hour, the doctor would be able to visualize the strain on his back over the given time period. The classifier is also able to identify back illnesses and with the two data sets we have used, we have achieved a 96% accuracy.

Challenges I ran into

Getting two accelerometers of the model given to us interface simultaneously- despite being limited by the number of ports available on the Arduino Uno.

Pull Down resistors

Pointer addressing Multiplexers using transistors(broke apart our previous project) but decided against since above method much faster

Embedded Systems are not built for writing data such as csv files into computers, but we figured out a way to read the data based on the script in MATLAB

Aligning two flex sensors into one and linearising it by mapping it as such

Processing raw data to values we can make sense of - like x, y, z into Gravity_x etc Running a regression model of these values and train a classifier

Extrapolates readings based on current data

Accomplishments that I'm proud of

96% accuracy from the 2 datasets we ran. Established threshold sensitivity to make very accurate models

What I learned

To not start devpost submission 10 minutes before

What's next for Asian Mom

Training the ML model to recognize different kinds of back illnesses and estimate time before they develop a back illness based on current posture record.

Built With

Share this project:
×

Updates