GPU-Powered Deep Learning Being Used to Speed Colon Cancer Diagnosis
Time is critical in cancer diagnosis. By spotting the disease early, and determining how fast it’s likely to spread, doctors can save lives.
Using GPU-powered deep learning, researchers at the Chinese University of Hong Kong have pushed the boundaries of cancer image analysis in a way that could one day save physicians — and patients — precious time.
The team’s work focused on colon cancer — the third most common cancer worldwide — recently took top honors at a challenge contest held at the Medical Image Computing and Computer Assisted Interventions (MICCAI) conference, the world’s leading conference on medical imaging.
Pathologists diagnose cancer by looking for abnormalities in tumor tissue and cells. The more abnormal, the more likely the cancer will grow and spread quickly. Traditionally, pathologists do this by examining tissue under a microscope. It’s a time-consuming process that’s open to error.
Determining Malignancy with Deep Learning
Using a TITAN X GPU, the research team was able to quickly train computers on a relatively small set of images of known abnormalities. The systems then used this training for segmenting individual glands from tissues to make it easier to distinguish individual cells, determine their size, shape and location relative to other cells. By calculating these measurements, pathologists can determine the likelihood of malignancy.
“Training with GPUs was 100 times faster than with CPUs,” said Hao Chen, a third-year Ph.D. student and member of the team that developed the solution. “That speed is going to become even more important as we advance our work.”
Deep learning uses complex neural networks to train computers to identify patterns and objects. It excels at problems like face detection and recognition, speech recognition and image classification. It’s already delivering better than humans for some tasks.
“This work is an advance in terms of a computer-aided diagnosis system,” said Pheng Ann Heng, chairman of the Computer Science Department at the Chinese University of Hong Kong and team lead.
Applications for Other Cancers
This competition involved only 165 images. Much more development and testing is needed to make the researchers’ work practical in the real world. If successful, this method could be applied to breast, lung and prostate cancer, which have certain similarities to colon cancer.
“We are really excited to see that the contest has already managed to push the boundaries of the state of the art, which was exactly the objective of organizing this contest,” said Dr Nasir Rajpoot, a contest organizer and head of the Bioimage Analysis Lab at the University of Warwick, U.K.