Go to Polygence Scholars page
Ritvik Suraparaju's cover illustration
Polygence Scholar2023
Ritvik Suraparaju's profile

Ritvik Suraparaju

Irvington High SchoolClass of 2023Fremont, California

About

Projects

  • "How can the latest advancements in Deep Learning (DL) technologies enable an AI-based computer program to help facilitate early breast cancer diagnosis, especially towards underserved rural towns in developing countries where radiologists are usually untrained to the point where a second opinion is almost always desired?" with mentor Shreya (Feb. 10, 2023)

Project Portfolio

How can the latest advancements in Deep Learning (DL) technologies enable an AI-based computer program to help facilitate early breast cancer diagnosis, especially towards underserved rural towns in developing countries where radiologists are usually untrained to the point where a second opinion is almost always desired?

Started Nov. 9, 2021

Abstract or project description

Research Question

How can the latest advancements in Deep Learning (DL) technologies enable an AI-based computer program to help facilitate early breast cancer diagnosis, especially towards underserved rural towns in developing countries where radiologists are usually untrained to the point where a second opinion is almost always desired?

Project Abstract:

Breast carcinoma is one of the most invasive tumors that accounts for cancer deaths in females. Early diagnosis is key and extremely critical , so treatment can be started early. It is proven that early diagnosis increases survivability and decreases the overall mortality rate. Recently, Artificial Intelligence (AI) technology has been explored to assist in the breast cancer diagnosis. Since most medical records are being uploaded digitally these days, it provides the opportunity to utilize Deep Learning to make more informed decisions. There are various imaging modalities that radiologists, oncologists and pathologists typically use for breast cancer diagnosis - Digital Mammography (DM), Ultrasound, Magnetic Resource Imaging (MRI) and Histopathology (HP) being the most prominent of them. However most of this work has been done with pre-processed data, in which image data used for model training is either a series of specific attribute values that represent each specific raw image or in many other cases the image modality chosen was Histopathology slides. While these techniques serve as an experimentation on various algorithms involved in the technology, they don’t offer any means where a radiologist or patient can use the raw mammogram image to directly get an output real-time.

This project aims at using the raw mammogram images as a data set in training the Deep Learning based data model. The big picture for this project is to make AI-based technology to serve as an assistive tool in having patients or radiologists to take second opinion on breast cancer diagnosis by simply supplying the mammogram image to a website that runs the AI-program in the backend. Such a cyber-radiologist tool becomes very useful especially in underserved rural towns in developing countries where there are a huge number of misdiagnoses and underdiagnosis of breast cancer resulting in breast cancer being unnoticed until it’s too late. We will be using the MIAS data set[1] that contains 322 images. Since the data set is not large enough, data augmentation will be performed to enhance the dataset and build better models for improvement in performance metrics like accuracy and AUC-ROC curve etc. The architecture involves Convolutional Neural Network (CNN) in feature extraction phase and in Neural Network in classification phase. Currently the scope of the project is to be able to classify cases into either benign or malignant, with visualization tools provided to localize the cancer region, that can help doctors verify the results.

References: https://www.kaggle.com/kmader/mias-mammography (Mias dataset)