An end-to-end process for cancer identification from images of lung tissue.

摘要

The purpose of this study was to develop a prototype for a non-interactive, computer-based second opinion diagnostic tool that could read microscope images of lung tissue and classify the tissue sample as normal or cancerous. This problem can be broken down into three areas: segmentation, feature extraction and measurement, and classification. This study introduces a kernel-based extension of fuzzy c-means to provide a coarse initial segmentation, with heuristically-based mechanisms to improve the accuracy of the segmentation. The segmented image is then processed to extract and quantify features. Finally, the measured features are used by a Support Vector Machine (SVM) to classify the tissue sample. The performance of this approach was tested using a training data set of 83 images, and an independent validation data set of 79 images, all collected at the Moffitt Cancer Center and Research Institute. These images represent a wide variety of normal lung tissue samples, as well as multiple types of lung cancer. The process developed was able to correctly classify 67% of the images in the validation set with a Receiver Operating Characteristic (ROC) curve area (AZ) of up to 0.704. When using only the images of normal tissue and a single type of cancer, the process was able to achieve up to 81% accuracy with a ROC AZ of 0.806.