Clinical practice has revealed ambiguities in PI-RADS v2.1 scoring, but a limited number of studies are available that validate the interreader and intrareader reproducibility of the mpMRI PI-RADS lexicon. We decomposed the PI-RADS rules into a set of common data elements to evaluate the inter- and intraobserver agreement in assessing the individual features included in the PI-RADS lexicon. Six radiologists (three highly experienced, three less experienced) in two sessions independently read thirty-two lesions in the peripheral and transition zone using the structured reporting tool, blinded to clinical MRI indication. The highest agreement between radiologists was observed for the abnormality detection, the evaluation of the type of signal intensity, and the characteristic of benign prostatic hyperplasia. Moderate agreement was reported for dynamic contrast-enhanced images. This resulted in a decrease in abnormality detection (PA = 76.5%) and enhancement indication (PA = 77.3%). The lowest agreement was observed for highly subjective features: shape, signal intensity level, and type of lesion margins. The results indicate the limitations of the PI-RADS v2.1 lexicon in relation to interreader and intrareader reproducibility. We have demonstrated that it is possible to develop structured reporting systems standardized according to the PI-RADS lexicon.

Background: Prostate cancer is one of the most common cancers worldwide. Currently, convolution neural networks (CNNs) are achieving remarkable success in various computer vision tasks, and in medical imaging research. Various CNN architectures and methodologies have been applied in the field of prostate cancer diagnosis. In this work, we evaluate the impact of the adaptation of a state-of-the-art CNN architecture on domain knowledge related to problems in the diagnosis of prostate cancer. The architecture of the final CNN model was optimised on the basis of the Prostate Imaging Reporting and Data System (PI-RADS) standard, which is currently the best available indicator in the acquisition, interpretation, and reporting of prostate multi-parametric magnetic resonance imaging (mpMRI) examinations. Methods: A dataset containing 330 suspicious findings identified using mpMRI was used. Two CNN models were subjected to comparative analysis. Both implement the concept of decision-level fusion for mpMRI data, providing a separate network for each multi-parametric series. The first model implements a simple fusion of multi-parametric features to formulate the final decision. The architecture of the second model reflects the diagnostic pathway of PI-RADS methodology, using information about a lesion’s primary anatomic location within the prostate gland. Both networks were experimentally tuned to successfully classify prostate cancer changes. Results: The optimised knowledge-encoded model achieved slightly better classification results compared with the traditional model architecture (AUC = 0.84 vs. AUC = 0.82). We found the proposed model to achieve convergence significantly faster. Conclusions: The final knowledge-encoded CNN model provided more stable learning performance and faster convergence to optimal diagnostic accuracy. The results fail to demonstrate that PI-RADS-based modelling of CNN architecture can significantly improve performance of prostate cancer recognition using mpMRI.

In this paper we propose a method, which supports determination of clinical significance of prostate lesions. The goal was to differentiate clinical significance of abnormalities according to the indicated lesion location. The clinical context of the reported research is based on multiparemetric Magnetic Resonanse Imaging (mpMRI) obtained from the PROSTATEx SPIE challenge database. The PROSTATEx training data set contains 203 mpMRI cases with 330 lesions located in various prostate zones. The proposed method is based on textural and statistical features of imaged tissue extracted from normalized T2-weighted (transaxial, sagittal and coronal), diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) and K-trans modalities. In addition patients' characteristics (age, weight, height) and lesion location were used. The impact of the modeled regions of interests of dimensions (2D or 3D) and size was also investigated. Standard classification method (K-nearest neighbors) has been applied in order to determine the most discriminative set of image and clinical features. The presented method achieved Area under Receiver Operating Curve (AUC) value equal to 0.92. Results indicate an interaction that affects lesion classification results of an area of an image used for feature extraction (lesion margin) and used texture features. The recommended set of features can be utilized in prostate cancer diagnosis and differentiation from benign prostate conditions.

A reproducible deep learning framework is presented for surface metrology to predict surface texture parameters together with their reported standard uncertainties. Using a multi-instrument dataset spanning tactile and optical systems, we jointly address measurement system type classification and regression of key surface parameters—arithmetic mean roughness (Ra), mean peak-to-valley roughness (Rz), and total roundness deviation (RONt)—alongside their reported standard uncertainties. Uncertainty is modelled via quantile and heteroscedastic regression heads, with post hoc conformal calibration used to obtain calibrated prediction intervals. On a held-out test set, high fidelity was achieved by single-target regressors (coefficients of determination: Ra 0.9824, Rz 0.9847, RONt 0.9918), with two uncertainty targets also well modelled (standard uncertainty of Ra 0.9899, standard uncertainty of Rz 0.9955); the standard uncertainty of RONt remained more difficult to learn (0.4934). The classifier reached 92.85% accuracy, and probability calibration was essentially unchanged after temperature scaling (expected calibration error 0.00504 → 0.00503 on the test split). Negative transfer was observed for naive multi-output trunks, with single-target models performing better. These results provide calibrated predictions suitable for informing instrument selection and acceptance decisions in metrological workflows.