A group of researchers from Tsinghua University in China were recently named first-place winners of a Kaggle’s Data Science Bowl for successfully developing algorithms that accurately detect signs of lung cancer in low-dose CT scans.
The winners of the $500,000 prize had a twofold strategy: first identify nodules and then diagnose cancer. Their approach also involved a neural network and an additional data set. “We think that explicitly dividing this problem into two stages is critical, which seems also to be what human experts would do,” Zhe Li, a member of the winning group, told the MIT Technology Review.
Contestants were only allowed to review 2,000 high-resolution lung scans provided by the (NCI) to create algorithms geared toward improving lung cancer screening methods. According to Keyvan Farahani, program director at the NCI, current protocol for identifying lung cancer isn’t always accurate. Participants aimed to decrease the false positive rate in CT scans by developing algorithms that detect cancerous lesions in the lungs.
More than 10,000 people submitted over 18,000 algorithms in this year’s Bowl. Second-place winners were two Dutch software and machine learning engineers and third place was Team Aidence, a medical deep learning company that focuses on image interpretation.
Although there are no plans for clinicians to implement the algorithms into diagnosis procedures, there’s the potential for medical imaging to incorporate the technology into future machine learning technology.