Key points are not available for this paper at this time.
The aim of this study is to find an appropriate group of spectral features for a multi-class machine learning (ML) architecture that provides a fast, effective, and accurate discrimination of chondrosarcoma nuclei irradiated with carbon ions (CI), X-rays (XR), and non-irradiated (REF), and that can be used for comparative assessment of the effects caused by the two types of radiation. The candidates were four groups of scalar and vector features drawn from the hyperspectral images (HIS) of the SW1353 chondrosarcoma line whose selection was guided by criteria including features relevance, cross entropy loss, robustness, running times, memory usage, and overall classification accuracy. The best results are obtained for six scalar features restricted to 415–460 nm spectral window. The associated ML classifier is subsequently used to compare the specific changes provoked by X-rays and carbon-ions by the help of the distance metric in the probability space where sorting is formally considered the action of prediction operators. We introduce the concept of confusion domain as a suitable approach able to disclose different shifts of the spectral features of SW1353 cells by type of irradiation. The findings exhibit consistency across all sections of the confusion domain thus supporting the conclusions on the different effects of radiation type. The results indicate that REF class is closer to XR class than to CI class and consequently X-rays produce weaker changes to the relevant spectral properties compared to carbon-ions. The biological resilience of the cells’ nuclei is higher against X-rays than against carbon-ions irradiation. The practical significance of our study lies in demonstrating the possibility to automate the decision-making process in measuring the amplitude of the ionizing radiation effects at single cell level, using HSI of unstained nuclei, with potential impact in radiotherapy of radioresistant cells.
Irimescu et al. (Thu,) studied this question.