Early non-invasive detection of breast cancer using exhaled breath and urine analysis. (1st May 2018)

• Record Type: Journal Article
• Authors: Herman-Saffar, Or; Boger, Zvi; Libson, Shai; Lieberman, David; Gonen, Raphael; Zeiri, Yehuda
• Is Part Of: Computers in biology and medicine. Volume 96(2018)
• Page Start: 227
• Page End: 232
• Publication Date: 2018-05-01
• Subjects: Exhaled breath -- Urine -- Breast cancer diagnosis -- Artificial neural networks; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2018.04.002
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: The main focus of this pilot study is to develop a statistical approach that is suitable to model data obtained by different detection methods. The methods used in this study examine the possibility to detect early breast cancer (BC) by exhaled breath and urine samples analysis. Exhaled breath samples were collected from 48 breast cancer patients and 45 healthy women that served as a control group. Urine samples were collected from 37 patients who were diagnosed with breast cancer based on physical or mammography tests prior to any surgery, and from 36 healthy women. Two commercial electronic noses (ENs) were used for the exhaled breath analysis. Urine samples were analyzed using Gas-Chromatography Mass-Spectrometry (GC-MS). Statistical analysis of results is based on an artificial neural network (ANN) obtained following feature extraction and feature selection processes. The model obtained allows classification of breast cancer patients with an accuracy of 95.2% ± 7.7% using data of one EN, and an accuracy of 85% for the other EN and for urine samples. The developed statistical analysis method enables accurate classification of patients as healthy or with BC based on simple non-invasive exhaled breath and a urine sample analysis. This study demonstrates that available commercial ENs can be used, provided that the data analysis is carried out using an appropriate scheme. Graphical abstract: Highlights: We demonstrate that commercial electronic noses (ENs) and GC-MS can be used to detect initial stages of breast cancer. A simple non-invasive sample collection and novel data analysis yield reliable results for breast cancer diagnosis. Preprocessed data used to develop an model based on Artificial Neural Networks (ANN) yield very poor results. Feature extraction leads to reduces number of feature and improves signal to noise ratio. Feature extraction improves ANN based models but they are still far from being satisfactory. The feature selection leads to models that yield very reliable results when used as simple non-invasive screening process.