CNNcon: A Quantitative Imaging Tool for Lung CT Image Feature Analysis

Abstract

Background: Lung CT scans are widely used for lung cancer screening and diagnosis. Current research focuses on quantitative analytics (radiomics) to improve screening and detection accuracy. However there are very limited numbers of portable software tools for automatic lung CT image analysis. Results: Here we build a Docker container, CNNcon, as a quantitative imaging tool for analyzing lung CT image features. CNNcon is developed from our recently published algorithm for nodule analysis, based on convolutional neural networks (CNN). When provided with a list of the centroid coordinates of regions of interest (ROI) in a volumetric CT study containing potential lung nodules, CNNcon can automatically generate highly accurate malignancy prediction of each ROI. CNNcon can also generate a vector of image features of each ROI, to facilitate further analyses by combining image features and other clinical features. As a Docker container, CNNcon is portable to various computer systems, convenient to install, and easy to use. CNNcon was tested on different computer systems and generated identical results. Conclusions: We anticipate that CNNcon will be a useful tool and broadly acceptable to the research community interested in quantitative image analysis.

COVID-19 Lung CT Image Segmentation Using Deep Learning Methods: UNET Vs. SegNET

“…Tumors have been direct targets for DL-assisted segmentation of medical images. In [4], a lung cancer screening tool was implemented using DL structures aiming to lower the false positive rate in lung cancer screening with low-dose CT scans. Also, in [5], researchers attempted to segment brain tumors from MRI images with a hybrid network of UNET and SegNet, reaching an accuracy of 0.99.…”