Project

#	Title	Team Members	TA	Documents	Sponsor
24	Machine Learning Enabled Wearable Stethoscope	Erlis Kllogjri Natalia Migdal Samuel Felder	Hershel Rege	design_document0.pdf final_paper0.pdf final_paper0.pdf other0.pdf proposal0.pdf
	There have been many recent advances around using several machine learning methods to detect and identify abnormalities in heart beat and lung breath audio. We are proposing a wearable system (to be worn around the chest) which will record audio, and analyze it, looking for abnormalities in the heart beats or lung sounds. This would provide a significant improvement in care for people at risk of issues with heart or lungs because it would be equivalent to having the attention of a doctor at all times. Use cases include examples such as firefighters (who have high rates of heart defects, and are also at risk of smoke inhalation on the job), hospital patients coming out of heart or lung surgery, or people who have history of issues with heart or lungs. The sensor would comprise of a sensitive microphone (and associated DSP circuitry), which would pick up the sounds from the heart or lungs. This would then get sent to the next sub-unit, which would process the audio. Depending on the complexity of either implementation (and the time and resource limitations) we could either use a microprocessor to implement a k-NN or a CNN algorithm to identify the sounds, or process through a hardware implementation of the k-NN or CNN. Once a worrisome sound is identified, it is communicated to the relevant party. In the case of hospital patients, it would be communicated to the doctor and nurses station. In the case of at home care, it would be communicated to doctor and emergency services. Finally in the case of emergency personnel (firefighters) it would be communicated to the captain and other emergency personnel. The communication would be implemented through RF. This has many advantages; it can be integrated with existing medical pager system, and since it only communicates when there is an issue it does not always need to be activated (and in the use cases it would only need to activate a handful of times), saving in power requirements.

Featured Project

The idea of this project would be to improve upon the 395 project ‘Smart Frisbee’ done by a group that included James Younce. The improvements would be to create a wristband with low power / short range RF capabilities that would be able to transmit a user ID to the frisbee, allowing the frisbee to know what player is holding it. Furthermore, the PCB from the 395 course would be used as a point of reference, but significantly redesigned in order to introduce the transceiver, a high accuracy GPS module, and any other parts that could be modified to decrease power consumption. The frisbee’s current sensors are a GPS module, and an MPU 6050, which houses an accelerometer and gyroscope.

The software of the system on the frisbee would be redesigned and optimized to record various statistics as well as improve gameplay tracking features for teams and individual players. These statistics could be player specific events such as the number of throws, number of catches, longest throw, fastest throw, most goals, etc.

The new hardware would improve the frisbee’s ability to properly moderate gameplay and improve “housekeeping”, such as ensuring that an interception by the other team in the end zone would not be counted as a score. Further improvements would be seen on the software side, as the frisbee in it’s current iteration will score as long as the frisbee was thrown over the endzone, and the only way to eliminate false goals is to press a button within a 10 second window after the goal.