fairfolio
Saved Projects

A Deep Learning Approach for Analyzing Tumor Histopathological Image Slides to Predict Breast Cancer

CSEF · 2018 Mathematical Sciences Honorable_mention Award

Ramya Ayyagari · Amador Valley High School, Pleasanton, Alameda County

Overview

Objectives/Goals The goal of this project was to create a highly-accurate, automated detection platform that would be able to predict breast cancer by classifying histology slides from biopsies. The method must group images into benign and malignant subtypes, and should be integrated on a cloud server to eventually be easily accessible to any pathologist. This would greatly lower false positive/negative rates in diagnosis. Methods/Materials Using Keras and TensorFlow, I built two Convolutional Neural Networks (CNN), deep learning algorithms that were trained on the BACH (Breast Cancer Histology) dataset, to classify histology slides into binary (carcinoma, non-carcinoma) and multiple (normal, benign, in-situ, invasive) categories. The CNNs were then fine-tuned and run various times on an Amazon (AWS) GPU instance to ensure reproducibility of results. To achieve better performance, the CNNs were trained on top of deep, pre-trained neural networks, such as VGG16 and InceptionResNetV2. The CNN algorithms were then integrated into an automated detection platform that I built on the cloud. Results The fine-tuned binary classification CNN, when coupled with the pre-trained layers of the model InceptionResNetV2, achieved an accuracy of 95.0%. This is a significant improvement on the state-of-the-art accuracy for a similar dataset, 83.3%, indicating that my model surpassed a current gold standard. The multi-class classification CNN with VGG16 pre-trained layers achieved an accuracy of 83.75%, surpassing the state-of-the-art accuracy for a similar dataset, 77.8%. Conclusions/Discussion The contributions of this project include providing a clinically applicable platform to accurately diagnose breast cancer via tumor slides. This is important because the tedious work of classifying these slides manually is time-consuming, inaccurate, and costly. In the future, pathologists can use this to quickly validate their diagnoses, and patients will be able to receive results within seconds. Additionally, this project provides a deep learning framework that can be applied to other medical computer vision problems.

Summary statement

I built a deep learning platform that can predict breast cancer with a 95% accuracy quickly and efficiently on a cloud server.

Help received

I used the publicly available BACH dataset to train my model, and the open source platforms Keras and TensorFlow. I designed and performed all experiments myself.

Awards (1)

  • Honorable Mention

Competition history

  • CSEF 2018 Mathematical Sciences · Entry S1502

Resources

Related projects

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

Browse more like this

Source: California Science & Engineering Fair public projects

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google