Differential privacy preserved federated learning for prognostic modeling in COVID-19 patients using large multi-institutional chest CT dataset.

BACKGROUND: Notwithstanding the encouraging results of previous studies reporting on the efficiency of deep learning (DL) in COVID-19 prognostication, clinical adoption of the developed methodology still needs to be improved. To overcome this limitation, we set out to predict the prognosis of a large multi-institutional cohort of patients with COVID-19 using a DL-based model. PURPOSE: This study aimed to evaluate the performance of deep privacy-preserving federated learning (DPFL) in predicting COVID-19 outcomes using chest CT images. METHODS: After applying inclusion and exclusion criteria, 3055 patients from 19 centers, including 1599 alive and 1456 deceased, were enrolled in this study. Data from all centers were split (randomly with stratification respective to each center and class) into a training/validation set (70%/10%) and a hold-out test set (20%). For the DL model, feature extraction was performed on 2D slices, and averaging was performed at the final layer to construct a 3D model for each scan. The DensNet model was used for feature extraction. The model was developed using centralized and FL approaches. For FL, we employed DPFL approaches. Membership inference attack was also evaluated in the FL strategy. For model evaluation, different metrics were reported in the hold-out test sets. In addition, models trained in two scenarios, centralized and FL, were compared using the DeLong test for statistical differences. RESULTS: The centralized model achieved an accuracy of 0.76, while the DPFL model had an accuracy of 0.75. Both the centralized and DPFL models achieved a specificity of 0.77. The centralized model achieved a sensitivity of 0.74, while the DPFL model had a sensitivity of 0.73. A mean AUC of 0.82 and 0.81 with 95% confidence intervals of (95% CI: 0.79-0.85) and (95% CI: 0.77-0.84) were achieved by the centralized model and the DPFL model, respectively. The DeLong test did not prove statistically significant differences between the two models (p-value = 0.98). The AUC values for the inference attacks fluctuate between 0.49 and 0.51, with an average of 0.50 ± 0.003 and 95% CI for the mean AUC of 0.500 to 0.501. CONCLUSION: The performance of the proposed model was comparable to centralized models while operating on large and heterogeneous multi-institutional datasets. In addition, the model was resistant to inference attacks, ensuring the privacy of shared data during the training process.

Deep Learning-based calculation of patient size and attenuation surrogates from localizer Image: Toward personalized chest CT protocol optimization.

PURPOSE: Extracting water equivalent diameter (DW), as a good descriptor of patient size, from the CT localizer before the spiral scan not only minimizes truncation errors due to the limited scan field-of-view but also enables prior size-specific dose estimation as well as scan protocol optimization. This study proposed a unified methodology to measure patient size, shape, and attenuation parameters from a 2D anterior-posterior localizer image using deep learning algorithms without the need for labor-intensive vendor-specific calibration procedures. METHODS: 3D CT chest images and 2D localizers were collected for 4005 patients. A modified U-NET architecture was trained to predict the 3D CT images from their corresponding localizer scans. The algorithm was tested on 648 and 138 external cases with fixed and variable table height positions. To evaluate the performance of the prediction model, structural similarity index measure (SSIM), body area, body contour, Dice index, and water equivalent diameter (DW) were calculated and compared between the predicted 3D CT images and the ground truth (GT) images in a slicewise manner. RESULTS: The average age of the patients included in this study (1827 male and 1554 female) was 53.8 ± 17.9 (18-120) years. The DW, tube current ,and CTDIvol measured on original axial images in the external 138 cases group were significantly larger than those of the external 648 cases (P < 0.05). The SSIM and Dice index calculated between the prediction and GT for body contour were 0.998 ± 0.001 and 0.950 ± 0.016, respectively. The average percentage error in the calculation of DW was 2.7 ± 3.5 %. The error in the DW calculation was more considerable in larger patients (p-value < 0.05). CONCLUSIONS: We developed a model to predict the patient size, shape, and attenuation factors slice-by-slice prior to spiral scanning. The model exhibited remarkable robustness to table height variations. The estimated parameters are helpful for patient dose reduction and protocol optimization.

COVID-19 prognostic modeling using CT radiomic features and machine learning algorithms: Analysis of a multi-institutional dataset of 14,339 patients.

BACKGROUND: We aimed to analyze the prognostic power of CT-based radiomics models using data of 14,339 COVID-19 patients. METHODS: Whole lung segmentations were performed automatically using a deep learning-based model to extract 107 intensity and texture radiomics features. We used four feature selection algorithms and seven classifiers. We evaluated the models using ten different splitting and cross-validation strategies, including non-harmonized and ComBat-harmonized datasets. The sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were reported. RESULTS: In the test dataset (4,301) consisting of CT and/or RT-PCR positive cases, AUC, sensitivity, and specificity of 0.83 ± 0.01 (CI95%: 0.81-0.85), 0.81, and 0.72, respectively, were obtained by ANOVA feature selector + Random Forest (RF) classifier. Similar results were achieved in RT-PCR-only positive test sets (3,644). In ComBat harmonized dataset, Relief feature selector + RF classifier resulted in the highest performance of AUC, reaching 0.83 ± 0.01 (CI95%: 0.81-0.85), with a sensitivity and specificity of 0.77 and 0.74, respectively. ComBat harmonization did not depict statistically significant improvement compared to a non-harmonized dataset. In leave-one-center-out, the combination of ANOVA feature selector and RF classifier resulted in the highest performance. CONCLUSION: Lung CT radiomics features can be used for robust prognostic modeling of COVID-19. The predictive power of the proposed CT radiomics model is more reliable when using a large multicentric heterogeneous dataset, and may be used prospectively in clinical setting to manage COVID-19 patients.

Effect of intra-abdominal administration of ascites fluid on postoperative peritoneal adhesion in rat model: A randomized controlled trail.

INTRODUCTION: Intra-abdominal adhesions are typically found after the most surgical procedures. Normally, most adhesions are asymptomatic; however, few individuals experience postoperative adhesion-related problems such as small bowel obstruction, pelvic pain, infertility, or other complications. We aimed to evaluate the preventive effect of the ascites fluid for postoperative peritoneal adhesions in rat models. MATERIAL AND METHODS: This experimental trial was conducted in Sixty Syrian male rat randomly assigned to six groups of 10 animals each as follows: control (group 1&4); normal saline (group 2&5): 2 mL of normal saline was poured into the peritoneal cavity; and case (group 3&6): 2 mL ascites fluid was poured into the peritoneal cavity. All animals in the six groups underwent laparotomy and measurable serosal injury were created with a standard technique. 10 and 30 days after initial surgery, the rats underwent another laparotomy in groups 1, 2, 3 and 4, 5, 6, respectively to assess macroscopic and microscopic adhesions, which were scored by an examiner who was blind to the animals̕ group assignment. Data analyzed by SPSS version 18, using the kruskal Wallis and Bonferroni-corrected Mann-Whitney U tests. P-values of less than 0.05 were considered significant. RESULTS: The mean scores of both microscopic and macroscopic adhesion were significantly different between all the groups (P < 0.05). Total macroscopic and microscopic adhesion scores were significantly lower in the ascites fluid treatment than in the control (P = 0.0001) or the normal saline (P < 0.001) group. There was no significant difference between adhesion intensity 10 and 30 days after laparotomy (P > 0.05). CONCLUSIONS: Ascites fluid can decrease the possibility of post-operative intraperitoneal adhesion formation.

Electroencephalography Pattern Variations During Motor Skill Acquisition.

The present study examined how motor skill acquisition affects electroencephalography patterns and compared short- and long-term electroencephalography variations. For this purpose, 17 volunteers with no history of disease, aged 18 to 22 years, attended seven training sessions every other day to practice a pursuit tracking motor skill. Electroencephalography brainwaves were recorded and analyzed on the first and last days within pre- and post-training intervals. The results showed a significant decrease in performance error and variability with practice over time. This progress slowed at the end of training, and there was no significant improvement in individual performance at the last session. In accordance with performance variations, some changes occurred in brainwaves. Specifically, θ power at Fz and α power at Cz increased on the last test day, compared with the first, while the coherence of α at Fz-T3 and Fz-Cz decreased. ß Coherence between Fz-Cz was significantly reduced from pre- to posttest. Based on these results, power changes seem to be more affected by long-term training, whereas coherence changes are sensitive to both short- and long-term training. Specifically, ß coherence at Fz-Cz was more influenced by short-term effects of training, whereas θ power at Fz, α power at Cz, and α coherence at Fz-T3 and Fz-Cz were affected by longer training.