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.

The importance of a correct positioning of the heart using IQ-SPECT system with multifocal collimators in myocardial perfusion imaging: a phantom study.

BACKGROUND: We have recently validated a quarter-time protocol in Myocardial Perfusion Imaging named IQ-SPECT, whose basic principle is to implement a multifocal collimator; However, in clinical practice, it may sometimes be difficult to center the heart in the region of highest magnification of the multifocal collimators (the so-called sweet spot). We therefore aimed to evaluate whether a heart mispositioning may affect results in MPI. METHODS: We simulated a rest study with an anthropomorphic phantom with an in vivo distribution of 400 MBq [(99m)Tc]tetrofosmin, with and without a transmural defect (TD). For each set of images, we performed 5 acquisitions, one with a correct centering and with other 4 degrees of mispositioning. Raw data and reconstructed images were evaluated qualitatively and quantitatively, including no corrections, correction for attenuation, for scatter or for both. We assessed polar plot uniformity, LV wall thickness, and TD and cavity contrast. RESULTS: Images obtained either with a correct heart centering or with mild misposition showed no differences, both qualitatively and quantitatively. Those obtained with major mispositioning differed in uniformity and TD contrast depending on correction parameters. CONCLUSION: This is the first study investigating how a heart mispositioning can affect diagnostic accuracy with IQ-SPECT system. Mild-to-moderate mispositioning (≤2.5 cm) is unlikely to significantly affect results.

IQ SPECT allows a significant reduction in administered dose and acquisition time for myocardial perfusion imaging: evidence from a phantom study.

UNLABELLED: We recently demonstrated in a clinical trial the ability of a new protocol, IQ SPECT, to acquire myocardial perfusion imaging (MPI) studies in a quarter of the time (12 s/view) of the standard protocol, with preserved diagnostic accuracy. We now aim to establish the lower limit of radioactivity that can be administered to patients and the minimum acquisition time in SPECT MPI using an IQ SPECT protocol, while preserving diagnostic accuracy. METHODS: An anthropomorphic cardiac phantom was used to acquire clinical rest scans with a simulated in vivo distribution of (99m)Tc-tetrofosmin at full dose (740 MBq) and at doses equal to 50%, 25%, and 18%. For each dose, 2 sets of images were acquired, with and without a transmural defect (TD). Variable acquisition times were also used for each dose. We analyzed raw data and reconstructed images, including no correction and correction for attenuation (AC), for scatter (SC), or for both (ACSC). Images were evaluated qualitatively and quantitatively in order to assess left ventricle (LV) wall thickness (full width at half maximum of the medial sections), TD, and cavity contrast in the LV wall. Data were compared across different acquisition times within the same dose and across doses with the same acquisition time. RESULTS: Images were visually scored as very-good quality except those acquired with 4 s/view or less at 100% dose and 6 s/view or less with 50%, 25%, or 18% dose, due to false-positive defects. LV wall thickness was not significantly different among all acquisitions. Cavity contrast remained unchanged within the same dose for all images and tended to be higher in AC and ACSC images. TD contrast remained unchanged within the same dose for all images. In SC and no-correction images, contrast was constant for all doses. AC images had significantly higher TD contrast values, and ACSC images showed a drop in TD contrast for a 50% dose. CONCLUSION: IQ SPECT effectively preserved both image quality and quantitative measurements with reduced acquisition time or administered dose in a phantom study. These findings suggest that approximately one eighth of the time, compared with standard protocols with a full dose, or a lower dose at an acquisition time of 12 s/view can be applied in MPI without the loss of diagnostic accuracy.