COPD Detection Algorithm for Use with Stethoscopes

Overview

COPD is the #3 global killer largely because of misdiagnosis rates in frontline medicine. The current gold- standard frontline diagnostic test is a questionnaire with a misdiagnosis rate of 65%, and misdiagnosis rates globally may be in excess of 90%; Failure to detect and treat patients early leads to poorer outcomes and higher healthcare costs. The aim of this project was to develop an algorithm that enables digital stethoscopes to detect COPD reliably. Stethoscopes are noninvasive and already in common use in frontline medicine today. 232 clinical stethoscope recordings were used. The recordings included healthy patients, patients with COPD, and patients with a variety of respiratory diseases. Each recording was reduced to four physical metrics: maximal Lyapunov exponent, correlation dimension, turbulence intensity and sample entropy. The metrics were then fitted to binary disease classification using logistic regression. The algorithm was evaluated based on sensitivity, specificity, p-values, and area under the ROC. In addition, a smaller training sample was selected randomly, refitted and evaluated against a test set of about 20%. The best result was achieved with 79 anterior site recordings. The algorithm was 94.9% sensitive and 95.0% specific; area under the ROC was 0.97. Using only these anterior recordings, a training set of 60 recordings was randomly selected and refitted to binary classification. The algorithm was then evaluated against remaining recordings; it was 77.8% sensitive and 88.9% specific. It was concluded that an effective COPD diagnostic can be created from stethoscope data that could significantly improve early identification of the disease. The intended application was frontline medicine, but the algorithm also proved more accurate than spirometry which is reported to have an accuracy rate of only 67%. Secure and Efficient Routing of Wireless Sensor Network Using Blockchain and Deep-Learning Based Algorithms Shraman Kar duPont Manual High School, Louisville, KY Industrial Io Tsystems have boomed due to recent innovations in wireless communication and digital electronics. An Industrial Io Tsystem consists of sensors/devices which connected with each other through a wireless network and collectively called a “Wireless Sensor Network”. These sensors provide real-time data on a scale of seconds and have been used in applications ranging from agriculture to entertainment. This system communicates data quickly with the base station through networks of their sensors. However, the open, distributed, and dynamic characteristics of these networks make them vulnerable to various types of attacks, thus seriously affecting their security and effectiveness. In 2021 alone, there were 155.8 million individuals around the world being affected by this. These attacks can have disastrous consequences and thus, it is paramount to find a solution to improve the efficiency and safety of wireless sensor networks. This research proposes a blockchain and reinforcement learning-based system to improve the security and efficiency of these networks. This system can prevent almost all attacks on the WSN compared with current system which is only able to prevent 80% of attacks and this system on average 45% more efficient than current systems.

Competition history

• JSHS 2022 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.