Project Title: Lung Cancer Detection with Deep Learning Convolutional Neural Networks
BASIS Advisor: Ms. Bhattacharya
Internship Location: Duke University
Onsite Mentor: Dr. Yiran Chen
Lung cancer is currently the deadliest disease in the world, killing 1.3 million people annually. Due to the large amount of work required to determine whether a patient has lung cancer, it would be extremely beneficial if doctors could use rapid-developing computer technology to help optimize the screening process of lung cancer. At Duke University, I will build an AlexNet and a VGG derived lung cancer detector with aggressive data augmentation, altering the last branch by transferring the original ability of classifying typical images to nodule images. I believe that my AlexNet and VGG models will classify a cancerous nodule correctly around 93% and 95% of the time respectively. In comparison, human doctors only correctly detect cancerous nodules 23% of the time. I will further improve my model through the AIFT algorithm, which continuously fine-tunes the CNN by incrementally enlarging the training dataset with newly annotated samples. Thus, I can reduce the number of labels without a significant sacrifice in accuracy. I hope that my research will be a convenient assistive tool in lung cancer diagnosis and change the world by saving millions of lives.
Learning Experiment #7 Results: On Friday April 5, I started training Learning Experiment #8. On Sunday April 7, I started training Learning Experiment #9 .
On Saturday March 30th, I started training Learning Experiment #5. On Monday April 1st, I started training Learning Experiment #6. On Wednesday April 3rd, I started training Learning Experiment #7.
On Thursday, I started training my first learning experiment. With the amount of data that I am training and testing, I decided to look up the best way to analyze and record my data. In the Youtube video attached below, it discuses the difference between a false positive, false negative, true positive, and true negative. […]
I was successfully able to convert the CT scans from a mhd to a jpg file format I generated a nodule image mask based on the position of a potential nodule (provided in the annotation file of the LUNA 16 Dataset). The region around the nodule is set to white while all other regions are left […]
I downloaded Data from LUNA 16: https://luna16.grand-challenge.org/data/. Since theGoogle Drive and Dropbox links were not working, I was forced to download using academic torrents. I had problems with torrents because never used them before so I decided to watch online tutorials. Another problem I faced this week was how to divide the data because all the […]
The AIFT algorithm (published in 2017) is a new, promising algorithm to reduce the number of labels required in the training process. It naturally integrates active learning and transfer learning into a single framework. AIFT continually fine-tunes the CNN by incrementally enlarging the training dataset with newly annotated samples. Thus, we can reduce the number […]
According to the syllabus I submitted in January, Week 3 of my senior project is supposed to be dedicated to continued research on the concepts directly related to the project. Sadly, this week I wasn’t able to get in as much work as I wanted to because I got a severe eye infection of Wednesday […]
Project Title: Lung Cancer Detection with Deep Learning Convolutional Neural Networks BASIS Advisor: Ms. Bhattacharya Mode of Daily Contact: Blog Goals: Lung cancer is currently the deadliest disease in the world, killing 1.3 million people annually. The main goal of my project is to better the understand the impact of how machine learning can help […]