Machine-learning-based classification of the histological subtype of non-small-cell lung cancer using MRI texture analysis. (April 2021)

• Record Type: Journal Article
• Authors: Bębas, Ewelina; Borowska, Marta; Derlatka, Marcin; Oczeretko, Edward; Hładuński, Marcin; Szumowski, Piotr; Mojsak, Małgorzata
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: PET/MRI -- Lung cancer -- Texture analysis -- Adenocarcinoma -- Squamous cell carcinoma -- Classification; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102446
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: PET / MR lung images with two types of cancer were used for the analysis. Various methods of image texture analysis were used. Several classification methods were used, including SVM, kNN, RF, deep learning. Division into adenocarcinoma and squamous cell carcinoma with 75.48 % efficiency. Abstract: Objective: Accurate differentiation of the histological sub-type of ono-small-cell lung cancer (NSCLC) at an early stage of diagnosis is crucial in choosing appropriate treatment option as soon as possible. Our study have been undertaken to classify two types of NSCLC – adenocarcinoma (ADC) and squamous cell carcinoma (SCC). Methods: PET / MR images from 44 patients with diagnosed adenocarcinoma (24 patients) and squamous cell carcinoma (20 patients) were used for the study. We managed to obtain 155 regions of interest with a size of 128 × 128 pixels, which were used for further analysis. 135 texture parameters were calculated, which were then used for classification using different types of classifiers. Results: The best results (75.48 %) were achieved using the Support Vector Machines (SVM) classifier and texture parameters histogram of oriented gradients (HOG) while obtaining the highest values of specificity and sensitivity. The other results of the classification are satisfactory to a large extent. Conclusion: The achieved results are satisfactory and give an opportunity to develop and improve the diagnostic process of non-small cell lung tumors at the imaging stage.