We investigate, by an extensive quality evaluation approach, performances and potential side effects introduced in Computed Tomography (CT) images by Deep Learning (DL) processing. Method: We selected two relevant processing steps, denoise and segmentation, implemented by two Convolutional Neural Networks (CNNs) models based on autoencoder architecture (encoder-decoder and UNet) and trained for the two tasks. In order to limit the number of uncontrolled variables, we designed a phantom containing cylindrical inserts of different sizes, filled with iodinated contrast media. A large CT image dataset was collected at different acquisition settings and two reconstruction algorithms. We characterized the CNNs behavior using metrics from the signal detection theory, radiological and conventional image quality parameters, and finally unconventional radiomic features analysis. Results: The UNet, due to the deeper architecture complexity, outperformed the shallower encoder-decoder in terms of conventional quality parameters and preserved spatial resolution. We also studied how the CNNs modify the noise texture by using radiomic analysis, identifying sensitive and insensitive features to the denoise processing. Conclusions: The proposed evaluation approach proved effective to accurately analyze and quantify the differences in CNNs behavior, in particular with regard to the alterations introduced in the processed images. Our results suggest that even a deeper and more complex network, which achieves good performances, is not necessarily a better network because it can modify texture features in an unwanted way.

The current framework of radiological protection of occupational exposed medical workers reduced the eye-lens equivalent dose limit from 150 to 20 mSv per year requiring an accurate dosimetric evaluation and an increase understanding of radiation induced effects on Lens cells considering the typical scenario of occupational exposed medical operators. Indeed, it is widely accepted that genomic damage of Lens epithelial cells (LEC) is a key mechanism of cataractogenesis. However, the relationship between apoptosis and cataractogenesis is still controversial. In this study biological and physical data are combined to improve the understanding of radiation induced effects on LEC. To characterize the occupational exposure of medical workers during angiographic procedures an INNOVA 4100 (General Electric Healthcare) equipment was used (scenario A). Additional experiments were conducted using a research tube (scenario B). For both scenarios, the frequencies of binucleated cells, micronuclei, p21-positive cells were assessed with different doses and dose rates. A Monte-Carlo study was conducted using a model for the photon generation with the X-ray tubes and with the Petri dishes considering the two different scenarios (A and B) to reproduce the experimental conditions and validate the irradiation setups to the cells. The simulation results have been tallied using the Monte Carlo code MCNP6. The spectral characteristics of the different X-ray beams have been estimated. All irradiated samples showed frequencies of micronuclei and p21-positive cells higher than the unirradiated controls. Differences in frequencies increased with the delivered dose measured with Gafchromic films XR-RV3. The spectrum incident on eye lens and Petri, as estimated with MCNP6, was in good agreement in the scenario A (confirming the experimental setup), while the mean energy spectrum was higher in the scenario B. Nevertheless, the response of LEC seemed mainly related to the measured absorbed dose. No effects on viability were detected. Our results support the hypothesis that apoptosis is not responsible for cataract induced by low doses of X-ray (i.e. 25 mGy) while the induction of transient p21 may interfere with the disassembly of the nuclear envelop in differentiating LEC, leading to cataract formation. Further studies are needed to better clarify the relationship we suggested between DNA damage, transient p21 induction and the inability of LEC enucleation.

. Purpose The aim of this work is the development and characterization of a model observer (MO) based on convolutional neural networks (CNNs), trained to mimic human observers in image evaluation in terms of detection and localization of low-contrast objects in CT scans acquired on a reference phantom. The final goal is automatic image quality evaluation and CT protocol optimization to fulfill the ALARA principle. Approach Preliminary work was carried out to collect localization confidence ratings of human observers for signal presence/absence from a dataset of 30,000 CT images acquired on a PolyMethyl MethAcrylate phantom containing inserts filled with iodinated contrast media at different concentrations. The collected data were used to generate the labels for the training of the artificial neural networks. We developed and compared two CNN architectures based respectively on Unet and MobileNetV2, specifically adapted to achieve the double tasks of classification and localization. The CNN evaluation was performed by computing the area under localization-ROC curve (LAUC) and accuracy metrics on the test dataset. Results The mean of absolute percentage error between the LAUC of the human observer and MO was found to be below 5% for the most significative test data subsets. An elevated inter-rater agreement was achieved in terms of S-statistics and other common statistical indices. Conclusions Very good agreement was measured between the human observer and MO, as well as between the performance of the two algorithms. Therefore, this work is highly supportive of the feasibility of employing CNN-MO combined with a specifically designed phantom for CT protocol optimization programs.

The 2013/59/Euratom Directive reduced the occupational exposure limits for the lens. Since it has become crucial to estimate the dose absorbed by the lens, we have studied the individual variability of exposed workers’ ocular conformations with respect to the data estimated from their personal dosimetry. The anterior eye conformations of 45 exposed workers were acquired using Scheimpflug imaging and classified according to their sight conditions (emmetropia, myopia or hypermetropia). Three eye models were computed, with two lens reconstructions, and implemented in an interventional radiology scenario using Monte Carlo code. The models were dosimetrically analysed by simulating setup A, a theoretical monoenergetic and isotropic photon source (10–150 keV) and setup B, a more realistic interventional setting with an angiographic x-ray unit (50, 75, 100 kV peak). Scheimpflug imaging provided an average anterior chamber depth of (6.4 ± 0.5) mm and a lens depth of (3.9 ± 0.3) mm, together with a reconstructed equatorial lens length of (7.1–10.1) mm. Using these data for model reconstruction, dose coefficients (DCs) were simulated for all ocular structures. Regardless of the eye model used, the DCs showed a similar trend with radiation energy, which highlighted that for the same energy and setup, no significant dependence on ocular morphology and workers’ visual conditions was observed. The maximum difference obtained did not exceed 1% for all eye models or structures analysed. Therefore, the individual variabilities of worker ocular anatomy do not require any additional correction, compared to the personal dosimetry data measured with a dedicated lens dosimeter. To estimate the dose absorbed by the other eye structures, it is, instead, essential to know the spectrum of the source that has generated the irradiation, since there are differences between monoenergetic sources and more realistic angiographic units.

No abstract