SIRIHAASA NALLAMOTHU

Machine Learning Researcher and Intern at Alectio. More research updates as project progresses.

Working with Dr.Jessie Chin at the University of Illinois at Urbana Champagin
Research on: Machine Learning, Data Science, Psychology and COVID-19 information.

Retinal and ocular diseases are a prevalent problem in today’s society with ophthalmic care being inaccessible. Over 200 million people worldwide suffer from retinal diseases relating to the fundus, and over 44 countries face the prevalent health problem of Trachoma which causes irreversible blindness. To solve these problems, I created six supervised binary classifications convolutional neural networks to diagnose AMD, Diabetic Retinopathy, Glaucoma, Hypertension, and Cataracts, each upwards of 90% accurate, and utilized model stacking for higher-quality processing and classification of fundus images. I also trained a neural network for the diagnosis of trachoma, via images of the inner surface of the eye-lid. These seven neural networks were trained off of a dataset with over 6,000 fundus images and 1,000 images of trachoma. When testing these neural networks with fundus with diabetic retinopathy, and passing it through the networks, all the networks except the ‘Normal vs. Diabetes’ models returned percentages around 1.9* 10^-5, proving that model stacking returns accurate results. I’m transferring these neural networks to raspberry pi with an Edge TPU for higher computational power, that will be connected to a camera, a touch-screen display [to show percentages], a light source, and a 20D ophthalmic lens, in order to create a cheap fundoscopy device. The device will serve as a cost-effective and accurate way for a user to take an image of the fundus or inner eyelid using the built-in camera and receive confidence percentages from the algorithms in a matter of minutes. View my research and work regarding for my developing prototype, Fovea, below.

LeukosCognosis is a 98% accurate Supervised Concolutional Nueral Network that can diagnose Acute Lymphoblastic Leukemia in a matter of seconds. I built this convolutional neural network through Tensorflow/Keras and Python and hope to deploy this in the medical field. I am currently writing a research paper for this project. Please watch my youtube video for more details. Every 9 minutes a person in the United States dies from Leukemia, or cancer of the blood and bone marrow. Essentially, Leukemia is a cancer of the blood and bone marrow and occurs when one’s body rapidly produces abnormal white blood cells, which decreases the body’s ability to fight infections. If Leukemia is not immediately diagnosed, as symptoms are often lucrative, this disease can act quickly. According to the Leukemia & Lymphoma Society, diagnosis and lab blood smear sample tests can take weeks to return, decreasing a patient’s ability and chances to survive. The need for a quick, effective, and proper diagnosis tool is urgent. In an effort to solve this problem, I created LeukosCognosis. LeukosCognosis is a Machine Learning Algorithm that can diagnose Acute Lymphoblastic Leukemia[ALL] on the spot with 97% accuracy. This algorithm can accurately diagnose Leukemia in seconds, a juxtaposition to the weeks that a lab Leukemia diagnosis may take. All the oncologist would have to do is take a microscopic image of the suspected white blood cell smear, and upload it to the model. Within seconds, the model will accurately diagnose the sample and predict the probability of Acute Lymphoblastic Leukemia. LeukosCognosis is unique as it provides a one-stop solution to long and often inaccurate lab tests and can be performed almost immediately. Once this model is deployed in the medical field, it will save time, money, and lives. I received funding for my research from various orginizations. View my research and work regarding LeukosCognosis below.

RiscusExplorato: A Machine Learning Algorithm to help aid in the Diagnosis of Covid-19 May 2020 – Present Project descriptionRiscusExplorato is a Machine Learning Algorithm to help aid the overall Diagnosis of Covid-19 via means of radiology and chest X-rays. The CDC has recognized extreme chest pain and shortness of breath as symptoms of coronavirus in this pandemic. These symptoms are often an indication of lung damage; using an X-ray can determine whether the patient is experiencing Viral Pneumonia, Covid-19 complications, or has Healthy lungs. Doctor's can Utilize RiscusExplorato’s classification to their advantage and use it as a factor in their diagnosis. All the radiologist has to do is take an image of the patient’s chest x-ray, upload it to the model, and receive a categorization or diagnosis. RiscusExplorato is essentially a Machine Learning algorithm that is 97% Accurate and can diagnose the Chest X-ray in a matter of seconds. Convolutional Neural Networks are a form of machine learning and entails many layers and pixel data. A Convolutional Network takes in an input image, which is then pixelated and analyzed through the Convolutional Layer. The Convolutional Layer essentially finds notable features in the image to help with classification, including edges, curves, and shapes. The pixelated image then goes through the pooling layer, which involves max pooling. This means that the pixelated image is broken down into groups and thoroughly analyzed for key features. The Neurons in the Dense layer essentially filter the key features and get it ‘ready’ for the final categorization of the image. The output layer provides a yes or no answer to the classification. RiscusExplorato will save time, money, and lives. RiscusExplorato has won first place in the Covid19 Equal Engineers International Challenge.

As humanity technologically advances and the human lifestyle surges forward, it is coming at an extremely steep cost-the environment. Gen Z has inherited a broken, pillaged Earth and it is our job to fix it before it's too late.Although there seems to be an infinite void of issues and challenges facing the environment, we can turn the current climate crisis around by all pitching in through a unified front. Through the right knowledge and resources, Gen Z can play a vital role in stopping Climate Change. This is where Environ comes into play. Environ harnesses the power of technology and the popularity of social media to enable users to make a difference in the environment and empowers them to influence and inspire others. Essentially, Environ makes it easy for anybody to be environmentally friendly. It allows users to set goals and make small climate-conscious lifestyle decisions that they can complete periodically. For example, users could set goals to replace plastic water bottles, bags, or straws with reusable ones, start composting, and even cut down on meat consumption. Environ also utilizes ‘influencing for good’, as users can create new ‘leafs’ or posts, to let others know what they have done! By providing achievable goals and simple solutions, along with harnessing the power of social media, Environ empowers anyone and everyone to make a change for the environment. I created this project with fellow Kode with a Klossy Scholaship winner. We received $2000 in seed funding from General Motors and Ashoka. We plan to expand our project with them. We also recived the finalist award for the Innovator Competition at the John PappJohn entrepreneurial center.

I created Support System as an 8th-9th grader, and won second place in the congressional app challenge, getting to meet congressman Darin LaHood. A passage from my research paper: Support System has 3 emergency screens. The first emergency screen is the general emergency screen. This screen can be used by anybody, even people who do not have an account with ​the Support System​. The general emergency screen benefits the user the most when they have an account. The screen can be used to access to a support hotline for suicide, depression, or rape, report sexual violence/assault, call or text an emergency contact (Only if you have an account), and even to text 9-1-1. This screen can also be used to send your location and a picture of your surroundings to an emergency contact. The second emergency screen is for those who have an account with Support System​. This screen is a natural disaster screen. The Natural disaster screen can be usedif there is a first aid need, a hailstorm, winter storm, dust storm, a Red Cross need, and more. This screen is essential for any natural disaster needs a person may have. The last emergency screen that is featured in this app is the Personal Emergency screen. This screen is exclusively for women with ​Support System​ Accounts. The Personal Emergency screen can be used to send automated and customized texts to an emergency contact. The user can customize texts for different personal emergencies. My app aims to improve the everyday lives of women and girls. After all, women everywhere need a ​Support System​.