Cancer is classified in two ways by the type of tissue in which it forms (histotype) and by the site of origin, or the location in the body where the cancer first formed. Now, researchers at Texas Tech University have developed a deep learning model that classifies cancer cells by type. Deep learning networks analyze images and classify cell types accurately and efficiently.
New research published in APLmachine learningAccording to the authors, it may be possible to further classify cancer severity.
“Cancer cells are highly heterogeneous, and recent studies suggest that specific cell subpopulations rather than whole cells are responsible for cancer metastasis,” said author Wei Lee, PhD, Texas. Associate Professor at the Institute of Technology) said. “Identifying cancer cell subpopulations is an important step in determining disease severity.”
“Cancer diagnosis is an important area for cancer recovery and survival, and many costly procedures are required to deliver adequate treatment,” the researchers wrote. “Machine learning (ML) approaches can help predict diagnoses from circulating tumor cells in liquid biopsies or from primary tumors in solid biopsies. This paper explores the use of deep convolutional neural networks to predict specific cancer cell lines as a label-free identification tool.”
“The problem with these complex and time-consuming technologies is the need for resources and effort that can be spent exploring different areas of cancer prevention and recovery,” said Carl, research assistant at Texas Tech University, author. Dr. Gardner said
“Our classification procedure does not consist of additional chemicals or biological solutions when taking pictures of cells,” says Gardner. “This is a ‘label-free’ method of identifying potential metastases.”
The team’s neural networks are also easy to use, efficient and automated. You input an image and the tool converts the data into probabilities. A result less than 0.5 he classifies the cancer into one cell type and a number greater than 0.5 indicates another cell type.
The tool was trained to optimize prediction accuracy using a series of images of two cancer cell lines. He reached an accuracy of over 94% across the datasets used in the study.
The authors aim to extend and generalize the model to include both single cells and clusters.
