Background: COVID assessment can be performed using the recently developed individual risk score (prediction of severe respiratory failure in hospitalized patients with SARS-COV2 infection, PREDI-CO score) based on High Resolution Computed Tomography. In this study, we evaluated the possibility of automatizing this estimation using semi-supervised AI-based Radiomics, leveraging the possibility of performing non-supervised segmentation of ground-glass areas. Methods: We collected 92 from patients treated in the IRCCS Sant’Orsola-Malpighi Policlinic and public databases; each lung was segmented using a pre-trained AI method; ground-glass opacity was identified using a novel, non-supervised approach; radiomic measurements were collected and used to predict clinically relevant scores, with particular focus on mortality and the PREDI-CO score. We compared the prediction obtained through different machine learning approaches. Results: All the methods obtained a well-balanced accuracy (70%) on the PREDI-CO score but did not obtain satisfying results on other clinical characteristics due to unbalance between the classes. Conclusions: Semi-supervised segmentation, implemented using a combination of non-supervised segmentation and feature extraction, seems to be a viable approach for patient stratification and could be leveraged to train more complex models. This would be useful in a high-demand situation similar to the current pandemic to support gold-standard segmentation for AI training.

Background: Benign renal tumors, such as renal oncocytoma (RO), can be erroneously diagnosed as malignant renal cell carcinomas (RCC), because of their similar imaging features. Computer-aided systems leveraging radiomic features can be used to better discriminate benign renal tumors from the malignant ones. The purpose of this work was to build a machine learning model to distinguish RO from clear cell RCC (ccRCC). Method: We collected CT images of 77 patients, with 30 cases of RO (39%) and 47 cases of ccRCC (61%). Radiomic features were extracted both from the tumor volumes identified by the clinicians and from the tumor’s zone of transition (ZOT). We used a genetic algorithm to perform feature selection, identifying the most descriptive set of features for the tumor classification. We built a decision tree classifier to distinguish between ROs and ccRCCs. We proposed two versions of the pipeline: in the first one, the feature selection was performed before the splitting of the data, while in the second one, the feature selection was performed after, i.e., on the training data only. We evaluated the efficiency of the two pipelines in cancer classification. Results: The ZOT features were found to be the most predictive by the genetic algorithm. The pipeline with the feature selection performed on the whole dataset obtained an average ROC AUC score of 0.87 ± 0.09. The second pipeline, in which the feature selection was performed on the training data only, obtained an average ROC AUC score of 0.62 ± 0.17. Conclusions: The obtained results confirm the efficiency of ZOT radiomic features in capturing the renal tumor characteristics. We showed that there is a significant difference in the performances of the two proposed pipelines, highlighting how some already published radiomic analyses could be too optimistic about the real generalization capabilities of the models.

Background: Microvascular invasion (MVI) is a necessary step in the metastatic evolution of hepatocellular carcinoma liver tumors. Predicting the onset of MVI in the initial stages of the tumors could improve patient survival and the quality of life. In this study, the possibility of using radiomic features to predict the presence/absence of MVI was evaluated. Methods: Multiphase contrast-enhanced computed tomography (CECT) images were collected from 49 patients, and the radiomic features were extracted from the tumor region and the zone of transition. The most-relevant features were selected; the dataset was balanced, and the presence/absence of MVI was classified. The dataset was split into training and test sets in three ways using cross-validation: the first applied feature selection and dataset balancing outside cross-validation; the second applied dataset balancing outside and feature selection inside; the third applied the entire pipeline inside the cross-validation procedure. Results: The features from the tumor areas on CECT showed both the portal and the arterial phases to be the most predictive. The three pipelines showed receiver operating characteristic area under the curve (ROC AUC) scores of 0.89, 0.84, and 0.61, respectively. Conclusions: The results obtained confirmed the efficiency of multiphase CECT and the ZOT in detecting MVI. The results showed a significant difference in the performance of the three pipelines, highlighting that a non-rigorous pipeline design could lead to model performance and generalization capabilities that are too optimistic.

The modern approach to treating rectal cancer, which involves total mesorectal excision directed by imaging assessments, has significantly enhanced patient outcomes. However, locally recurrent rectal cancer (LRRC) continues to be a significant clinical issue. Identifying LRRC through imaging is complex, due to the mismatch between fibrosis and inflammatory pelvic tissue. This work aimed to develop a machine learning model for predicting LRRC using radiomic features extracted from 18F-FDG Positron Emission Tomography/Computed Tomography (PET/CT). CT and PET images of PET/CT examinations were retrospectively collected from 44 patients, with 29 cases of recurrence (66 %) and 15 cases with no local recurrence (34 %). The whole analysis was conducted separately for CT and PET images to evaluate their different predictive power. Radiomic features were extracted from suspected lesion volumes identified by physicians and the most relevant radiomic features were selected to predict the presence or absence of LRRC. Feature selection was performed using a novel approach derived from gene expression analysis, based on the DNetPRO algorithm. The prediction was done using a Support Vector Classifier (SVC) with a 10-fold cross-validation. The efficiency of the pipeline in predicting LRRC was evaluated according to the sensitivity, specificity, Balanced Accuracy Score (BAS) and Matthews's Correlation Coefficient (MCC). CT features were found to be the most predictive, showing a sensitivity of 0.80, a specificity of 0.82, a BAS of 0.81 and an MCC of 0.61. PET features obtained a sensitivity of 0.93, a specificity of 0.61, a BAS of 0.77 and a MCC of 0.52. The combination of PET and CT features obtained a sensitivity of 0.80, a specificity of 0.75, a BAS of 0.77 and a MCC of 0.53. To the best of our knowledge, the DNetPRO algorithm was applied for the first time to medical image analysis and proved suitable for the selection of radiomic features with the highest predictive power, a crucial step in a radiomic pipeline. Our results confirmed the efficiency of radiomic features in predicting LRRC, with CT features outperforming PET features in identifying the characteristic texture of LRRC. The combination of both yielded no performance improvement.