Despite careful selection, the recurrence rate after upfront surgery for pancreatic adenocarcinoma can be very high. We aimed to construct and validate a model for the prediction of early distant recurrence (<12 months from index surgery) after upfront pancreaticoduodenectomy. After exclusions, 147 patients were retrospectively enrolled. Preoperative clinical and radiological (CT-based) data were systematically evaluated; moreover, 182 radiomics features (RFs) were extracted. Most significant RFs were selected using minimum redundancy, robustness against delineation uncertainty and an original machine learning bootstrap-based method. Patients were split into training (n = 94) and validation cohort (n = 53). Multivariable Cox regression analysis was first applied on the training cohort; the resulting prognostic index was then tested in the validation cohort. Clinical (serum level of CA19.9), radiological (necrosis), and radiomic (SurfAreaToVolumeRatio) features were significantly associated with the early resurge of distant recurrence. The model combining these three variables performed well in the training cohort (p = 0.0015, HR = 3.58, 95%CI = 1.98–6.71) and was then confirmed in the validation cohort (p = 0.0178, HR = 5.06, 95%CI = 1.75–14.58). The comparison of survival curves between low and high-risk patients showed a p-value <0.0001. Our model may help to better define resectability status, thus providing an actual aid for pancreatic adenocarcinoma patients’ management (upfront surgery vs. neoadjuvant chemotherapy). Independent validations are warranted.

Purpose. Different ML models were compared to predict toxicity in RT on a large cohort (n = 1314). Methods. The endpoint was RTOG G2/G3 acute toxicity, resulting in 204/1314 patients with the event. The dataset, including 25 clinical, anatomical, and dosimetric features, was split into 984 for training and 330 for internal tests. The dataset was standardized; features with a high p-value at univariate LR and with Spearman $\rho >0.8$ were excluded; synthesized data of the minority were generated to compensate for class imbalance. Twelve ML methods were considered. Model optimization and sequential backward selection were run to choose the best models with a parsimonious feature number. Finally, feature importance was derived for every model. Results. The model’s performance was compared on a training–test dataset over different metrics: the best performance model was LightGBM. Logistic regression with three variables (LR3) selected via bootstrapping showed performances similar to the best-performing models. The AUC of test data is slightly above 0.65 for the best models (highest value: 0.662 with LightGBM). Conclusions. No model performed the best for all metrics: more complex ML models had better performances; however, models with just three features showed performances comparable to the best models using many (n = 13–19) features.

The ongoing COVID-19 pandemic currently involves millions of people worldwide. Radiology plays an important role in the diagnosis and management of patients, and chest computed tomography (CT) is the most widely used imaging modality. An automatic method to characterize the lungs of COVID-19 patients based on individually optimized Hounsfield unit (HU) thresholds was developed and implemented. Lungs were considered as composed of three components—aerated, intermediate, and consolidated. Three methods based on analytic fit (Gaussian) and maximum gradient search (using polynomial and original data fits) were implemented. The methods were applied to a population of 166 patients scanned during the first wave of the pandemic. Preliminarily, the impact of the inter-scanner variability of the HU-density calibration curve was investigated. Results showed that inter-scanner variability was negligible. The median values of individual thresholds th1 (between aerated and intermediate components) were −768, −780, and −798 HU for the three methods, respectively. A significantly lower median value for th2 (between intermediate and consolidated components) was found for the maximum gradient on the data (−34 HU) compared to the other two methods (−114 and −87 HU). The maximum gradient on the data method was applied to quantify the three components in our population—the aerated, intermediate, and consolidation components showed median values of 793 ± 499 cc, 914 ± 291 cc, and 126 ± 111 cc, respectively, while the median value of the first peak was −853 ± 56 HU.