Background The Prostate Imaging Reporting and Data System, version 2.1 (PI-RADSv2.1) standardizes reporting of multiparametric MRI of the prostate. Assigned assessment categories are a risk stratification algorithm, higher categories indicate a higher probability of clinically significant cancer compared to lower categories. PI-RADSv2.1 does not define these probabilities numerically. We conduct a systematic review and meta-analysis to determine the cancer detection rates (CDR) of the PI-RADSv2.1 assessment categories on lesion level and patient level. Methods Two independent reviewers screen a systematic PubMed and Cochrane CENTRAL search for relevant articles (primary outcome: clinically significant cancer, index test: prostate MRI reading according to PI-RADSv2.1, reference standard: histopathology). We perform meta-analyses of proportions with random-effects models for the CDR of the PI-RADSv2.1 assessment categories for clinically significant cancer. We perform subgroup analysis according to lesion localization to test for differences of CDR between peripheral zone lesions and transition zone lesions. Results A total of 17 articles meet the inclusion criteria and data is independently extracted by two reviewers. Lesion level analysis includes 1946 lesions, patient level analysis includes 1268 patients. On lesion level analysis, CDR are 2% (95% confidence interval: 0–8%) for PI-RADS 1, 4% (1–9%) for PI-RADS 2, 20% (13–27%) for PI-RADS 3, 52% (43–61%) for PI-RADS 4, 89% (76–97%) for PI-RADS 5. On patient level analysis, CDR are 6% (0–20%) for PI-RADS 1, 9% (5–13%) for PI-RADS 2, 16% (7–27%) for PI-RADS 3, 59% (39–78%) for PI-RADS 4, 85% (73–94%) for PI-RADS 5. Higher categories are significantly associated with higher CDR (P < 0.001, univariate meta-regression), no systematic difference of CDR between peripheral zone lesions and transition zone lesions is identified in subgroup analysis. Conclusions Our estimates of CDR demonstrate that PI-RADSv2.1 stratifies lesions and patients as intended. Our results might serve as an initial evidence base to discuss management strategies linked to assessment categories.

Automatic prostate tumor segmentation is often unable to identify the lesion even if multi-parametric MRI data is used as input, and the segmentation output is difficult to verify due to the lack of clinically established ground truth images. In this work we use an explainable deep learning model to interpret the predictions of a convolutional neural network (CNN) for prostate tumor segmentation. The CNN uses a U-Net architecture which was trained on multi-parametric MRI data from 122 patients to automatically segment the prostate gland and prostate tumor lesions. In addition, co-registered ground truth data from whole mount histopathology images were available in 15 patients that were used as a test set during CNN testing. To be able to interpret the segmentation results of the CNN, heat maps were generated using the Gradient Weighted Class Activation Map (Grad-CAM) method. The CNN achieved a mean Dice Sorensen Coefficient 0.62 and 0.31 for the prostate gland and the tumor lesions -with the radiologist drawn ground truth and 0.32 with whole-mount histology ground truth for tumor lesions. Dice Sorensen Coefficient between CNN predictions and manual segmentations from MRI and histology data were not significantly different. In the prostate the Grad-CAM heat maps could differentiate between tumor and healthy prostate tissue, which indicates that the image information in the tumor was essential for the CNN segmentation.

Background: The aim of the study was to evaluate the role of different immunohistochemical and radiomics features in patients with small cell lung cancer (SCLC). Methods: Consecutive patients with histologically proven SCLC with limited (n = 47, 48%) or extensive disease (n = 51, 52%) treated with radiotherapy and chemotherapy at our department were included in the analysis. The expression of different immunohistochemical markers from the initial tissue biopsy, such as CD56, CD44, chromogranin A, synaptophysin, TTF-1, GLUT-1, Hif-1 a, PD-1, and PD-L1, and MIB-1/KI-67 as well as LDH und NSE from the initial blood sample were evaluated. H-scores were additionally generated for CD44, Hif-1a, and GLUT-1. A total of 72 computer tomography (CT) radiomics texture features from a homogenous subgroup (n = 31) of patients were correlated with the immunohistochemistry, the survival (OS), and the progression-free survival (PFS). Results: The median OS, calculated from diagnosis, was 21 months for patients with limited disease and 13 months for patients with extensive disease. The expression of synaptophysin correlated with a better OS (HR 0.546 95% CI 0.308-0.966, p = 0.03). The expression of TTF-1 (HR 0.286, 95% CI: 0.117-0.698, p = 0.006) and a lower GLUT-1 H-score (median = 50, HR: 0.511, 95% CI: 0.260-1.003, p = 0.05) correlated with a better PFS. Patients without chromogranin A expression had a higher risk for developing cerebral metastases (p = 0.02) and patients with PD 1 expression were at risk for developing metastases (p = 0.02). Our radiomics analysis did not reveal a single texture feature that correlated highly with OS or PFS. Correlation coefficients ranged between −0.48 and 0.39 for OS and between −0.46 and 0.38 for PFS. Gkika et al. Immunohistochemistry and Radiomics in SCLC Conclusions: The role of synaptophysin should be further evaluated as synaptophysin-negative patients might profit from treatment intensification. We report an, at most, moderate correlation of radiomics features with overall and progression free survival and no correlation with the expression of different immunohistochemical markers.

Automatic prostate tumor segmentation is often unable to identify the lesion even if in multi-parametric MRI data is used as input, and the segmentation output is difficult to verify due to the lack of clinically established ground truth images. In this work we use an explainable deep learning model to interpret the predictions of a convolutional neural network (CNN) for prostate tumor segmentation. The CNN uses a U-Net architecture which was trained on multi-parametric MRI data from 122 patients to automatically segment the prostate gland and prostate tumor lesions. In addition, co-registered ground truth data from whole mount histopathology images were available in 15 patients that were used as a test set during CNN testing. To be able to interpret the segmentation results of the CNN, heat maps were generated using the Gradient Weighted Class Activation Map (Grad-CAM) method. With the CNN a mean Dice Sorensen Coefficient for the prostate gland and the tumor lesions of 0.62 and 0.31 with the radiologist drawn ground truth and 0.32 with wholemount histology ground truth for tumor lesions could be achieved. Dice Sorensen Coefficient between CNN predictions and manual segmentations from MRI and histology data were not significantly different. In the prostate the Grad-CAM heat maps could differentiate between tumor and healthy prostate tissue, which indicates that the image information in the tumor was essential for the CNN segmentation.