Generative Electrical Impedance Tomography from Multichannel Stethoscope Sounds (2023 - present)

Electrical Impedance Tomography (EIT) reveals lung function and helps diagnose many diseases. Commercial EIT systems are expensive and bulky. Two thirds of the world’s population, especially in combat zones, does not even have access to basic radiology services such as X-rays, let alone EIT. However, stethoscopes are ubiquitous, cheap, and portable. This work proposed creating an AI method to generate EIT images from stethoscope sounds from multiple chest positions. A Variational Autoencoder (VAE) was created to map an EIT image to an embedding. Several convolutional neural networks (CNNs) and loss functions were tested to map mel spectrogram images of stethoscope sounds from multiple positions on the chest to EIT image embeddings, which were then passed through the VAE’s decoder to generate EIT images. To evaluate the accuracy of generated images, a CNN was trained to detect diseases from real EIT images. The accuracy of the generated EIT images using the disease classification CNN was compared to that from real EIT images. Not only was the generated data visually similar to real EIT images, but they also produced similar disease classifications as real images. The highest relative accuracy was 87.5% and average accuracy about 82% across five data splits. This is the first work to propose EIT image generation from stethoscope sounds. Future work includes improvements in networks and training, building a cheap stethoscope vest with synchronized capturing of sounds from multiple positions on the chest, and training a network to generate EIT images from the sounds.

NOTE:
I have filed a provisional patent at the USPTO for this invention
Publication in progress

Generating Chest X-Ray images from stethoscope sounds (2022 - present)

According to the World Health Organization, two thirds of the world does not have access to basic radiology services such as X-rays. Imaging the body with X-rays exposes the body to a small dose of ionizing radiation, which is associated with a slightly higher risk of developing cancer later on in life. Imagine a world where anyone, including patients without access to doctors or radiology centers, can get a chest X-ray without the need to travel long distances, and without the need to expose the chest to any ionizing radiation. This project allows one to synthesize an accurate chest X-ray image from stethoscope sounds directly. Today, many machine learning based methods are able to diagnose certain heart and lung diseases from stethoscope sounds or from chest X ray images. Therefore, it is plausible, as this project shows, that there is common information present in both stethoscope sounds and chest X-ray images that might allow us to convert stethoscope sounds to chest X-ray images. Today, there exist methods that can convert text to images and sounds to text (e.g. Google voice assistant, Apple’s Siri, etc). However, converting sounds directly to images has seen considerably less work. The longer term goal for this project is to pave the way for a cheap vest with microphones attached at several locations in the front and back of the chest, that can be worn by a patient. The microphones will capture sounds from the patient’s chest. This project demonstrates the feasibility of converting the captured stethoscope sounds to a chest X-ray image. Please note that the microphones can also capture not only sounds that humans can hear, but also infrasonic and ultrasonic sounds, so that additional useful information is presented to the neural networks that a human cannot exploit. I am currently collaborating with doctors and an X-ray clinic in India to collect patient data to train and validate better models.

NOTE:
I have filed for a patent at the USPTO for this invention (US-20240320874-A1)
Publication in progress

Materials to adsorb Methane (2021 - 2022)

Methane traps 35 times more heat than carbon dioxide, and continues to grow fast in our atmosphere. Recently, there has been increasing interest in reducing methane emissions and capturing the methane already present in the atmosphere. My project carried out a controlled experiment comparing the methane adsorption efficiency of zeolites, activated carbon, and curcumin. Two ideas to capture more methane were tested based on the facts that methane is slightly positively charged on the outside, and its diameter is 3.92 Å. An airtight chamber was constructed, and methane, generated by heating sodium acetate and sodium hydroxide (Dumas’ reaction), was passed through the materials. The drop in methane concentration after twenty minutes of exposure was measured with a sensor. This is the first work that shows that an electric charge on activated carbon has a significant effect on its methane adsorption, and that curcumin can adsorb a significant amount of methane.

CatCH4, a trash can to reduce methane from landfills (2021)

A high school friend and I presented a business idea at the CONRAD 2021 challenge. Our product is called catCH4, which is a device that will catch methane (CH4) at one of its sources: trash cans at home, restaurants, and other places handling food. Our device will be able to adsorb a significant amount of methane produced in a household trash can. We estimate that we can reduce global warming by up to 1-2% each year. As the organic waste in trash cans decompose anaerobically over time, methane is generated slowly and escapes into the atmosphere. We will pass methane through zeolite, a mineral that adsorbs methane. We got our idea for our project from a paper that proposed the use of large fans to blow air from the atmosphere through zeolites to capture methane and to heat the zeolite to release the methane as carbon dioxide. Our method is to use the same principle to catch methane being emitted from organic waste from trash cans at homes and restaurants. A solar powered fan will take the methane generated by the organic waste and push it through zeolite. We developed many ideas along the way, and we struggled to find the most effective method in which the methane would be adsorbed by the zeolite. There would be a tube through the lid, leading into a chamber with zeolite. We then realized that methane is lighter than air, and should rise to the top of the trash can naturally. After this development, we removed the tube and replaced it with a hole that will allow the entrance of methane. We will sandwich the zeolite in between two strips of gauze, which are big enough to let methane in, but small enough to prevent zeolite from falling through. Two small solar powered fans will be above the gauze strips, and a small solar panel will be on the outside of the lid.

Increasing the Bioavailability of Curcumin (2020 - present)

Curcumin, found in the spice turmeric, has been shown to kill cancer cells in vitro. However, curcumin has very low solubility in water, and therefore is very hard to be absorbed in the body. Curcumin dissolves in oil and alcohol, but curcumin molecules react with other molecules in the stomach and intestines. Nanoparticles up to size 100 nm have been shown to be better absorbed in the body, as they survive the digestive tract better. As they disintegrate slowly, individual molecules get freed up, and have a better chance of reaching cells in the body. Smaller nanoparticle sizes are better for absorption. There are many ways to produce curcumin nanoparticles. A recent article in the Nature journal showed that nanoparticles are formed when curcumin, dissolved in dimethylsulfoxide, is dropped into a fructose solution. Curcumin is used in cooking in countries such as India, where it is mixed with oil and then heated. The incidence of cancer is lower in India than in many countries. One possibility is that heating curcumin in oil forms curcumin nanoparticles. The goal of this project was to find out which method among Polyethylene Glycol (PEG) + Fructose, Dimethylsulfoxide + Fructose, heated oil, and oil + corn starch produces the smallest curcumin nanoparticles. PEG + Fructose showed the smallest nanoparticle size, followed by DMSO + Fructose, Oil heated 150 ⁰C, and Oil + starch. This research discovered many novel ways of creating curcumin nanoparticles, among which PEG + Fructose yielded the smallest sized nanoparticles.

NOTE:
Publication in progress