Search

Although preoperative biopsy of rectal cancer (RC) is an essential step for confirmation of diagnosis, it currently fails to provide prognostic information to the clinician beyond a rough estimation of tumour grade. In this study we used a risk classification to stratified patient in low-risk and high-risk patients in relation to the disease free survival and the overall survival using histopathological post-operative features. The purpose of this study was to evaluate if low-risk and high-risk RC can be distinguished using a CT-based radiomics model. We retrospectively reviewed the preoperative abdominal contrast-enhanced CT of 40 patients with RC. CT portal-venous phase was used for manual RC segmentation by two radiologists. Radiomics parameters were extracted by using PyRadiomics (3.0) software, which automatically obtained a total of 120 radiomics features. An operator-independent statistical hybrid method was adopted for the selection and reduction of features, while discriminant analysis was used to construct the predictive model. Postoperative histopathological report was used as reference standard. Receiver operating characteristics (ROC) and areas under the ROC curve (AUROC) were calculated to evaluate the diagnostic performance of the most dis-criminating selected parameters. Sensitivity, specificity, and accuracy were calculated. In our study cohort, the original_shape_Maximum3DDiameter and origi-nal_shape_MajorAxisLength demonstrated a good performance in the differentiation between preoperative degree of RC, with an AUROC of 0.680%, sensitivity of 74.02%, specificity of 73.45%, positive predictive value of 81.47%, and accuracy of 73.71%. In conclusion, this preliminary analysis showed statistically significant differences in radiomics features between low-risk and high-risk RC.

The purpose of this preliminary study was to evaluate the differences in Magnetic Resonance Imaging (MRI)-based radiomics analysis between cerebellopontine angle neurinomas and schwannomas originating from other locations in the neck spaces. Twenty-six patients with available MRI exams and head and neck schwannomas were included. Lesions were manually segmented on the precontrast and postcontrast T1 sequences. The radiomics features were extracted by using PyRadiomics software, and a total of 120 radiomics features were obtained from each segmented tumor volume. An operator-independent hybrid descriptive‐inferential method was adopted for the selection and reduction of the features, while discriminant analysis was used to construct the predictive model. On precontrast T1 images, the original_glcm_InverseVariance demonstrated a good performance with an area under the receiver operating characteristic (AUROC) of 0.756 (95% C.I. 0.532–0.979; p = 0.026). On postcontrast T1 images, the original_glcm_Idmn provided a good diagnostic performance with an AUROC of 0.779 (95% C.I. 0.572–0.987; p = 0.014). In conclusion, this preliminary analysis showed statistically significant differences in radiomics features between cerebellopontine angle neurinomas and schwannomas of other locations in the neck spaces.

SARS-CoV-2 epidemics has resulted in an unprecedented global health crisis causing a high number of deaths with pneumonia being the most common manifestation. Chest CT is the best imaging modality to identify pulmonary involvement, but unfortunately there are no pathognomonic features for COVID-19 pneumonia, since many other infectious and non-infectious diseases may cause similar alterations. The adoption of artificial intelligence in biomedical imaging has the potential to revolutionize the identification, management, and the patient’s outcome. If adequately validated, it could be used as a support with predictive and prognostic purposes in symptomatic patients but also as a screening test in asymptomatic patients in COVID-19 epidemics. Some studies have already shown the potential adoption of artificial intelligence for detection of COVID-19 infection, or even to differentiate from community-acquired pneumonia, but at present artificial intelligence cannot routinely applied for COVID-19 due to several limitations. This book chapter will first revise the basics of radiomics with a short practical and easy guide for radiologists; then, the main radiological findings of COVID-19 pneumonia will be presented with the most relevant information that are assessed to evaluate extent of the disease; finally, the main current literature on potential clinical application of radiomics and artificial intelligence for COVID-19 will be presented together with limitations and perspectives.

Magnetic Resonance Imaging-based prostate segmentation is an essential task for adaptive radiotherapy and for radiomics studies whose purpose is to identify associations between imaging features and patient outcomes. Because manual delineation is a time-consuming task, we present three deep-learning (DL) approaches, namely UNet, efficient neural network (ENet), and efficient residual factorized convNet (ERFNet), whose aim is to tackle the fully-automated, real-time, and 3D delineation process of the prostate gland on T2-weighted MRI. While UNet is used in many biomedical image delineation applications, ENet and ERFNet are mainly applied in self-driving cars to compensate for limited hardware availability while still achieving accurate segmentation. We apply these models to a limited set of 85 manual prostate segmentations using the k-fold validation strategy and the Tversky loss function and we compare their results. We find that ENet and UNet are more accurate than ERFNet, with ENet much faster than UNet. Specifically, ENet obtains a dice similarity coefficient of 90.89% and a segmentation time of about 6 s using central processing unit (CPU) hardware to simulate real clinical conditions where graphics processing unit (GPU) is not always available. In conclusion, ENet could be efficiently applied for prostate delineation even in small image training datasets with potential benefit for patient management personalization.