Advait developed a smartphone app prototype called Heart NN that lets people check their heart sounds.
HeartNN: A High-Accuracy Neural Network for Cardiac AuscultationView Poster
“Cardiac disease is extremely common, yet often goes undetected,” Advait says. He wondered if people with heart problems could live longer with the help of artificial intelligence, specifically machine learning. “Doctors manually classify heart sounds as normal or abnormal using stethoscopes,” Advait says. “But what if machines could do the same?” He had coded his first machine learning project last year, so he wanted to apply the same concept to this health problem. Advait created a smartphone app that lets people check their heart sounds.
Tactics and Results
Advait collected audio recordings of heart sounds labelled as normal or abnormal from a publicly available dataset. He then converted the recordings into visual images that show how the sounds change over time, called spectrograms. Using these, Advait designed computer software that can classify visualized heart sounds as normal or abnormal. After testing the software on recordings that it had never analyzed before, it ended up being almost 98% accurate. Advait put the software to use in a smartphone app called HeartNN. A user simply uploads an audio recording of their heart sounds to be examined by the software. After getting analyzed, the app shows the probability of the heart sounds being normal. If something seems off, the user can follow up with a doctor. “This app empowers people to perform initial cardiac auscultation [listening to sounds from the heart or other organs] from the comfort of their homes,” Advait says.
Beyond the Project
Adding steps to clean incoming data, like getting rid of background noise from the recordings, would further improve prediction accuracy, Advait says. He would then like many users and healthcare professionals to test the app before making it available for free in app stores.
“I like playing video games,” Advait says. “Minecraft gives me access to a world where I can build what I want.” He also enjoys the game GeoGuessr, where he must figure out the location of a randomly-shown place by examining various landmarks. On the musical side, Advait plays the mridangam, a percussion instrument that originates from India. One day, Advait would like to be a computer scientist. “A career as a computer scientist will give me a lot of opportunities to showcase my capabilities and also develop technologies that can practically impact the world around me,” he says.