A hybrid machine learning/deep learning COVID-19 severity predictive model from CT images and clinical data.

COVID-19 clinical presentation and prognosis are highly variable, ranging from asymptomatic and paucisymptomatic cases to acute respiratory distress syndrome and multi-organ involvement. We developed a hybrid machine learning/deep learning model to classify patients in two outcome categories, non-ICU and ICU (intensive care admission or death), using 558 patients admitted in a northern Italy hospital in February/May of 2020. A fully 3D patient-level CNN classifier on baseline CT images is used as feature extractor. Features extracted, alongside with laboratory and clinical data, are fed for selection in a Boruta algorithm with SHAP game theoretical values. A classifier is built on the reduced feature space using CatBoost gradient boosting algorithm and reaching a probabilistic AUC of 0.949 on holdout test set. The model aims to provide clinical decision support to medical doctors, with the probability score of belonging to an outcome class and with case-based SHAP interpretation of features importance.

Post Surgical Management of WHO Grade II Meningiomas: Our Experience, the Role of Gamma Knife and a Literature Review.

Purpose: Grade II meningiomas are rarer than Grade I, and when operated on, bear a higher risk of local recurrence, with a 5-year progression free survival (PFS) ranging from 59 to 90%. Radiotherapy (RT) or radiosurgery, such as Gamma Knife radiosurgery (GKRS) can reduce the risk of relapse in patients with residual disease, even if their role, particularly after gross total resection (GTR), is still under debate. Main goal of this study was to compare the outcomes of different post-surgical management of grade II meningiomas, grouped by degree of surgical removal (Simpson Grade); next in order we wanted to define the role of GKRS for the treatment of residual disease or relapse. Methods: from November 2016 to November 2020 all patients harboring grade II meningiomas, were divided into three groups, based on post-surgical management: (1) wait and see, (2) conventional adjuvant radiotherapy and (3) stereotactic GKRS radiosurgery. Relapse rate and PFS were registered at the time of last follow up and results were classified as stable, recurrence next to or distant from the surgical cavity. In the second part of the study we collected data of all patients who underwent GKRS in our Centers from November 2017 to November 2020. Results: A total of 37 patients were recruited, including seven patients with multiple meningiomas. Out of 47 meningiomas, 33 (70.2%) were followed with a wait and see strategy, six (12.7%) were treated with adjuvant radiotherapy, and 8 patients (17.0%) with adjuvant GKRS. Follow up data were available for 43 (91.4%) meningiomas. Within the wait and see group, recurrence rates differed based on Simpson grades, lower recurrence rates being observed in three Simpson I cases (30%) compared to twelve relapses (60%) in patients with Simpson grade II/III. Finally, out of the 24 meningiomas undergoing GKRS (8 residual and 16 recurrence), 21 remained stable at follow up. Conclusions: Gross total resection (GTR) Simpson II and III have a significantly worse outcome as compared to Simpson I. The absence of adjuvant treatment leads to significant worsening of the disease progression curve. Adjuvant radiotherapy, especially GKRS, provides good local control of the disease and should be considered as an adjuvant treatment in all cases where Simpson I resection is not possible.

F-formation detection: individuating free-standing conversational groups in images.

Detection of groups of interacting people is a very interesting and useful task in many modern technologies, with application fields spanning from video-surveillance to social robotics. In this paper we first furnish a rigorous definition of group considering the background of the social sciences: this allows us to specify many kinds of group, so far neglected in the Computer Vision literature. On top of this taxonomy we present a detailed state of the art on the group detection algorithms. Then, as a main contribution, we present a brand new method for the automatic detection of groups in still images, which is based on a graph-cuts framework for clustering individuals; in particular, we are able to codify in a computational sense the sociological definition of F-formation, that is very useful to encode a group having only proxemic information: position and orientation of people. We call the proposed method Graph-Cuts for F-formation (GCFF). We show how GCFF definitely outperforms all the state of the art methods in terms of different accuracy measures (some of them are brand new), demonstrating also a strong robustness to noise and versatility in recognizing groups of various cardinality.

Imaging characteristics evaluation of a digital radiography system.

PURPOSE: To evaluate the physical imaging characteristics of an indirect digital radiography system used for general radiography. MATERIAL AND METHODS: The performance of the two 41x41 cm2 CsI:Tl/a-Si flat-panel detectors of a GE-Revolution XR/d digital radiography system was evaluated. Signal uniformity, dose linearity, pre-sampling and expectation Modulation Transfer Function (MTF), Noise Power Spectrum (NPS) and Detective Quantum Efficiency (DQE) were measured at 70 kVp, varying the incident dose levels, according to IEC 62220-1. The effects of anti-scatter grid on NPS were also studied. RESULTS: Both detectors have a limiting spatial resolution of 2.5 lp/mm set by the pixel pitch. Raw images demonstrated fairly good uniformity, excellent repeatability and linearity. Without the grid in the anode-cathode direction, for an 8 mGy incident dose, both pre-sampling and EMTF were greater than 0.5 and 0.2 for spatial frequencies of 1 and 2 lp/mm, respectively. At the same dose and frequencies, for one detector, DQE was greater than 0.61 and 0.20; for the other one, DQE was greater than 0.41 and 0.18. DISCUSSION AND CONCLUSIONS: The two detector panels showed different DQE curves. Namely, the table detector DQE was excellent at low dose levels but its performance appeared to degrade with increasing doses, both in mean values and in shape, whereas the wall stand detector DQE appeared to depend less strongly on dose. In any case, the DQE values obtained from this study were higher than those reported in the literature for storage phosphor Computed Radiography systems.

Sentinel lymph node as a new marker for therapeutic planning in breast cancer patients.

BACKGROUND AND OBJECTIVES: Literature review suggests that the sentinel lymph node (sN) represents a reliable predictor of axillary lymph node status in breast cancer patients; however, some important issues, such as the optimisation of the technique for the intraoperative identification of the sN, the role of intraoperative frozen section examination of the sN, and the clinical implications of sN metastasis as regards the surgical management of the axilla, still require further confirmation. The authors aimed (1) to assess the feasibility of sN identification with a combined approach (vital blue dye lymphatic mapping and radioguided surgery, RGS) and the specific contribution of either techniques to the detection of the sN, (2) to determine the accuracy and usefulness of intraoperative frozen section examination of the sN in order to perform a one-stage surgical procedure, and (3) to define how the sN might modulate the therapeutic planning in different stages of disease. MATERIALS AND METHODS: From October 1997 to June 2001, 334 patients with early-stage (T(1-2) N(0) M(0)) invasive mammary carcinoma underwent sN biopsy; the average age of patients was 61.5 years (range, 39-75 years). In a subset of 153 patients, both vital blue dye (Patent Blue-V) lymphatic mapping and RGS were used to identify the sN, and the relative contribution of each of the two techniques was assessed. RESULTS: In the whole group, the sN was identified in 326 of 334 patients (97.6%), and 105 of 326 patients (37.3%) had positive axillary lymph nodes (pN+). In 9 of 105 pN+ patients, the definitive histologic examination of the sN did not show metastases but these were detected in non-sN, thus giving an 8.6% false-negative rate, a negative predictive value of 94.5% (156/165), and an accuracy of 96.5% (252/261). As regards the specific contribution of the two different techniques used in the identification of the sN, the detection rate was 73.8% (113/153) with Patent Blue-V alone, 94.1% (144/153) with RGS alone, and 98.7% (151/153) with Patent Blue-V combined with RGS (P < 0.001). Noteworthy, whenever the sN was identified, the prediction of axillary lymph node status was remarkably similar (93-95% sensitivity; 100% specificity; 95-97% negative predictive value, and 97-98% accuracy) whichever of the three procedures was adopted (Patent Blue-V alone, RGS alone, or combined Patent Blue-V and RGS). Intraoperative frozen section examination was performed in 261 patients, who had at least one sN identified, out of 267 patients who underwent complete axillary dissection; 170 patients had histologically negative sN (i.o. sN-) and 91 patients histologically positive sN (i.o. sN+). All 91 i.o. sN+ were confirmed by definitive histology, whereas in 14 of 170 i.o. sN- patients (8.2%) metastases were detected at definitive histology. As regards the correlation between the size of sN metastasis, the primary tumour size, and the status of non-sN in the axilla, micrometastases were detected at final histology in 23 patients and macrometastases in 82 patients. When only micrometastases were detected, the sN was the exclusive site of nodal metastasis in 20 of 23 patients (86.9%) while in 3 patients with tumour size larger than 10 mm micrometastases were detected also in non-sN. Macrometastases were never detected in pT(1a) breast cancer patients; the sN was the exclusive site of these metastases in 30 patients (36.6%), while in 52 patients (63.4%) there were metastases both in sN and non-sN. CONCLUSIONS: Sentinel lymphadenectomy can better be accomplished when both procedures (lymphatic mapping with vital blue dye and RGS) are used, because of the significantly higher sN detection rate, although the prediction of axillary lymph node status remains remarkably similar whichever method is used. The intraoperative frozen section examination proved to be rather accurate in predicting the actual pathologic status of the sN, with a negative predictive value of 91.8%; in 35% of patients it allowed sN biopsy and axillary dissection to be performed in a one-stage surgical procedure. Finally, specific clinical and histopathologic features of the primary tumour and sN might be used to tailor the loco-regional and systemic treatment in different clinical settings, such as in ductal carcinoma in-situ (DCIS), early-stage invasive breast cancer, and patients with large breast cancer undergoing neo-adjuvant CT for breast-saving surgery as well as elderly patients with operable breast cancer.