Publication date: 2023

DOI: 10.1007/978-3-031-35311-6_35

The article considers the use of deep neural networks for the recognition of lungs inhomogeneity from computed tomography (CT) scans, as well as the determination of the amount of inhomogeneity, which will help to diagnose respiratory diseases at an early stage. The recognition process is performed in several steps: lung segmentation on a CT slice, segmentation of inhomogeneous areas on the corresponding CT slice, pattern recognition, determination of the volume of inhomogeneity and generation of a report with a detailed description of the slices. The article describes the architecture of the neural network and the process of its training to identify inhomogeneous areas on a CT scan of the respiratory organs, as well as the developed algorithm for calculating the volume of inhomogeneity. To implement the recognition stage, a lightweight convolutional neural network (CNN) is used. The developed approach provides for finding areas containing inhomogeneous areas on scan slices, highlighting them in the image and determining the volume of inhomogeneity. Datasets from the public database MosMedData and NSCLC were used as data. The IOU and Dice metrics on the test data were 0.82 and 0.92, respectively. The software is implemented in the Python 3.6 programming language using the Django 2.2 framework and is cross-platform. For machine learning algorithms, PyTorch 1.9.0 packages were used. An analysis of the effectiveness of the results obtained showed a recognition accuracy of 91% according to the Dice metric. #CSOC1120. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Издатель: Springer Science and Business Media Deutschland GmbH

Тип: Article