Automatic Generation of Interpretable Lung Cancer Scoring Models from Chest X-Ray Images.
Year: 2020
Type: preprint
Abstract: Lung cancer is the leading cause of cancer death worldwide with early\ndetection being the key to a positive patient prognosis. Although a multitude\nof studies have demonstrated that machine learning, and particularly deep\nlearning, techniques are effective at automatically diagnosing lung cancer,\nthese techniques have yet to be clinically approved and adopted by the medical\ncommunity. Most research in this field is focused on the narrow task of nodule\ndetection to provide an artificial radiological second reading. We instead\nfocus on extracting, from chest X-ray images, a wider range of pathologies\nassociated with lung cancer using a computer vision model trained on a large\ndataset. We then find the set of best fit decision trees against an\nindependent, smaller dataset for which lung cancer malignancy metadata is\nprovided. For this small inferencing dataset, our best model achieves\nsensitivity and specificity of 85% and 75% respectively with a positive\npredictive value of 85% which is comparable to the performance of human\nradiologists. Furthermore, the decision trees created by this method may be\nconsidered as a starting point for refinement by medical experts into\nclinically usable multi-variate lung cancer scoring and diagnostic models.
Source: arXiv (Cornell University)
Authors: Michael J. Horry, Subrata Chakraborty, Biswajeet Pradhan, Manoranjan Paul, Douglas Pinto Sampaio Gomes, Anwaar Ulhaq
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3112952182