image segmentation

 Chest computed tomography (CT) scan is one of the most common tools used for the diagnosis of patients with coronavirus disease 2019 (COVID-19). While segmentation of COVID-19 lung lesions by radiologists can be time-consuming, the application of advanced deep learning techniques for automated segmentation can be a promising step toward the management of this infection and similar diseases in the future.

 This study aimed to evaluate the performance and generalizability of deep learning-based models for the automated segmentation of COVID-19 lung lesions.

 Four datasets (2 private and 2 public) were used in this study. The first and second private datasets included 297 (147 healthy and 150 COVID-19 cases) and 82 COVID-19 subjects. The public datasets included the COVID19-P20 (20 COVID-19 cases from 2 centers) and the MosMedData datasets (50 COVID-19 patients from a single center). Model comparisons were made based on the Dice similarity coefficient (DSC), receiver operating characteristic (ROC) curve, and area under the curve (AUC). The predicted CT severity scores by the model were compared with those of radiologists by measuring the Pearson’s correlation coefficients (PCC). Also, DSC was used to compare the inter-rater agreement of the model and expert against that of 2 experts on an unseen dataset. Finally, the generalizability of the model was evaluated, and a simple calibration strategy was proposed.

 The VGG16-UNet model showed the best performance across both private datasets, with a DSC of 84.23% ± 1.73% on the first private dataset and 56.61% ± 1.48% on the second private dataset. Similar results were obtained on public datasets, with a DSC of 60.10% ± 2.34% on the COVID19-P20 dataset and 66.28% ± 2.80% on a combined dataset of COVID19-P20 and MosMedData. The predicted CT severity scores of the model were compared against those of radiologists and were found to be 0.89 and 0.85 on the first private dataset and 0.77 and 0.74 on the second private dataset for the right and left lungs, respectively. Moreover, the model trained on the first private dataset was examined on the second private dataset and compared against the radiologist, which revealed a performance gap of 5.74% based on DSCs. A calibration strategy was employed to reduce this gap to 0.53%.

 The results demonstrated the potential of the proposed model in localizing COVID-19 lesions on CT scans across multiple datasets; its accuracy competed with the radiologists and could assist them in diagnostic and treatment procedures. The effect of model calibration on the performance of an unseen dataset was also reported, increasing the DSC by more than 5%.

Prostate-specific membrane antigen (PSMA) has been demonstrated as a promising tool for specific imaging of prostate cancer (PCa) via positron emission tomography-computed tomography (PET/CT) scanning. Radiation treatment planning (RTP) based on 68Ga-PSMA PET/CT scanning can also lead to some decision modifications.  The specific goal of this comparative study is to show how 68Ga-PSMA PET/CT images can influence the target volume delineation (TVD) and normal tissue radiation dose for PCa RTP, and to compare gross tumor volumes (GTVs) delineated using various strategies for 68Ga-PSMA PET-based image segmentation techniques. This study consisted of eleven 68Ga-PSMA PET/CT images related to patients affected with locally advanced PCa. Four strategies also included manual segmentation techniques, a 2.5 standardized uptake value (SUV) cutoff (SUV=2.5), as well as a fixed threshold of 40% and 50% of the maximum signal intensity (SUV=%40 SUVmax and SUV=%50 SUVmax) for 68Ga-PSMA PET-based segmentation techniques to delineate GTVPET. Two treatment planning were accordingly generated for each patient based on manual GTVPET and CT-only. The GTV was statistically and significantly smaller for PET/CT-derived volumes (9.39 vs. 77.98 cm3 for CT alone) (p<0.002). There was no significant difference in volumes of GTV2.5 and GTV40% with GTVman (p=0.11) although we observed a significant difference in volumes of GTV50% with GTVman (p=0.02). Mean bladder dose (MBD), V50 of rectum, and mean femoral dose (MFD) for PET/CT plans were significantly lower than CT-only (22.36 vs. 46.55 Gy; p=0.004), (33% vs. 67.82%; p=0.000), and (28.01 vs. 37.12Gy; p=0.013); respectively. The contribution of hybrid modalities of PSMA-PET/CT can be useful for detailed target volume planning and reduce radiation exposure to organs at risk. Using molecular images in RTP also demonstrates the biological volume of GTV so that it will not be left out of the field to cause recurrent tumor.

  In order to evaluate the growth rate of lung cancer, pulmonary nodule segmentation is an essential and crucial step. Segmentation of juxta-pleural pulmonary nodule in CT scans, especially small size ones, is still a challenge. To better support the following radiomics analysis, this study aims to propose and develop a novel segmentation method for small-size juxta-pleural pulmonary nodules. In this study, we investigated and developed a novel approach based on transition region thresholding and chain code analysis to segment juxta-pleural pulmonary nodules. First, we cropped the region of interest (ROI) from the lung CT scans, and enhanced the nodule regions by using an anisotropic diffusion algorithm. Second, to extract the foreground pixels (including the attached chest wall) from ROIs, we applied an adaptive segmentation process by incorporating a threshold segmentation method with transition region analysis. Third, we smoothed the lung contour by using iterative weighted averaging algorithm. Then, we utilized chain code analysis to repair lung parenchyma boundaries. Finally, we obtained the segmentation result by overlapping the extracted foreground with the repaired lung parenchyma mask. To validate the performance of the proposed segmentation approach, we selected 50 juxta-pleural nodules with diameter ranges from 5 mm to 10 mm from Lung Image Database Consortium (LIDC) database. Compared with the ground truth generated by radiologists, we achieved an average overlap rate of 76.93% ± 0.06 with a false positive rate of 13.09% ± 0.09. After comparing and analyzing the segmentation results, we found that our approach outperformed the method reported in other literature. The experimental results demonstrated that our new method is an effective approach to segment small-size juxta-pleural pulmonary nodules accurately.

Identification of surgical instruments in laparoscopic video images has several biomedical applications. While several methods have been proposed for accurate detection of surgical instruments, the accuracy of these methods is still challenged high complexity of the laparoscopic video images. This paper introduces a Surgical Instrument Detection Framework (SIDF) for accurate identification of surgical instruments in complex laparoscopic video frames.

Based on the Generalized Near-Set Theory, a novel image segmentation algorithm, termed Generalized Near-Set Theory-based Image Segmentation Algorithm (GNSTISA) was developed. According to SIDF, first GNSTISA is executed to segment the laparoscopic images. Next, the segments generated by GNSTISA are filtered based on their color and texture. The remaining segments would then indicate surgical instruments.

Using the laparoscopic videos of varicocele surgeries obtained from Hasheminezhad Kidney Center, the performance of GNSTISA was compared with previous image segmentation methods. The results showed that GNSTISA outperforms the earlier algorithms in term of accurate segmentation of laparoscopic images. Moreover, the accuracy of SIDF in identifying the surgical instruments was found superior to that of other methods.

SIDF eliminates the limitations of previous image segmentation methods, and can be used for precise identification of surgical instrument detection.