Computer Vision / Video Analytics

AI Helps Doctors Diagnose Lung Cancer with 97 Percent Accuracy

Sep 18, 2018
By Nefi Alarcon
Discuss (0)

AI-Generated Summary
  • Researchers at New York University developed a convolutional neural network based on the Inception v3 model that identifies two common types of lung cancer from histopathology slides with accuracy comparable to pathologists.
  • The model achieved ~0.99 AUC for distinguishing normal versus tumor tissue and 0.97 AUC for differentiating lung cancer types using whole-slide images.
  • Training used 3,000 images from public datasets with NVIDIA Tesla GPUs and the cuDNN-accelerated TensorFlow framework.
  • The study was published in Nature Medicine and the code is available on GitHub.

Next Steps

Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more

According to the American Cancer Society, over 200,000 people in the United States are diagnosed with lung cancer every year. To help with early detection, researchers at the New York University school developed a deep learning algorithm that can identify two common types of lung cancer with human accuracy.
“The purpose of this study was to develop a deep-learning model for the automatic analysis of tumor slides using publicly available whole-slide images and to subsequently test our models on independent cohorts collected at our institution,” the researchers stated in their paper.
The NYU team based their convolutional neural network on the open source Inception v3 image recognition model, which has been successfully used to detect skin cancer and diabetic retinopathy.
Using NVIDIA Tesla GPUs with the cuDNN-accelerated TensorFlow deep learning framework, the researchers re-trained the Inception convolutional neural network on 3000 images from two separate datasets.
The researchers then evaluated their dataset on whole slide images from separate datasets. They also enlisted three independent pathologists to classify the slides.

Data and strategy. a, Number of whole-slide images per class. b, Strategy for training. (b, i), Images of lung cancer tissues were first downloaded from the Genomic Data Commons database; (b, ii), slides were then separated into a training (70%), a validation (15%) and a test set (15%); (b, iii), slides were tiled by nonoverlapping 512- × 512-pixel windows, omitting those with over 50% background; (b, iv), the Inception v3 architecture was used and partially or fully re-trained using the training and validation tiles; (b, v), classifications were performed on tiles from an independent test set, and the results were finally aggregated per slide to extract the heatmaps and the AUC statistics. c, Size distribution of the images widths (gray) and heights (black). d, Distribution of the number of tiles per slide.

“Our study demonstrates that convolutional neural networks, such as Google’s inception v3, can be used to assist in the diagnosis of lung cancer from histopathology slides: it almost unambiguously classified normal versus tumor tissues (~0.99 AUC) and distinguishes lung cancer types with high accuracy (0.97 AUC), reaching sensitivity and specificity comparable to that of a pathologist,” the team stated on their paper.
The research was published this week in Nature Medicine, and the code was also published on GitHub.
Read more>

Discuss (0)

Tags

Computer Vision / Video Analytics | cuDNN | News | Machine Learning & Artificial Intelligence | Medical Imaging

About the Authors

Nefi Alarcon
About Nefi Alarcon
Nefi Alarcon is a senior executive communications manager on NVIDIA's leadership team. He has years of media relations and communication experience, and has previously worked at Google, Mozilla, and CNN. He received his bachelor's degree in Journalism from George Washington University.

Comments