Automated prostate multi-regional segmentation in magnetic resonance using fully convolutional neural networks.

OBJECTIVE: Automatic MR imaging segmentation of the prostate provides relevant clinical benefits for prostate cancer evaluation such as calculation of automated PSA density and other critical imaging biomarkers. Further, automated T2-weighted image segmentation of central-transition zone (CZ-TZ), peripheral zone (PZ), and seminal vesicle (SV) can help to evaluate clinically significant cancer following the PI-RADS v2.1 guidelines. Therefore, the main objective of this work was to develop a robust and reproducible CNN-based automatic prostate multi-regional segmentation model using an intercontinental cohort of prostate MRI. METHODS: A heterogeneous database of 243 T2-weighted prostate studies from 7 countries and 10 machines of 3 different vendors, with the CZ-TZ, PZ, and SV regions manually delineated by two experienced radiologists (ground truth), was used to train (n = 123) and test (n = 120) a U-Net-based model with deep supervision using a cyclical learning rate. The performance of the model was evaluated by means of dice similarity coefficient (DSC), among others. Segmentation results with a DSC above 0.7 were considered accurate. RESULTS: The proposed method obtained a DSC of 0.88 ± 0.01, 0.85 ± 0.02, 0.72 ± 0.02, and 0.72 ± 0.02 for the prostate gland, CZ-TZ, PZ, and SV respectively in the 120 studies of the test set when comparing the predicted segmentations with the ground truth. No statistically significant differences were found in the results obtained between manufacturers or continents. CONCLUSION: Prostate multi-regional T2-weighted MR images automatic segmentation can be accurately achieved by U-Net like CNN, generalizable in a highly variable clinical environment with different equipment, acquisition configurations, and population. KEY POINTS: • Deep learning techniques allows the accurate segmentation of the prostate in three different regions on MR T2w images. • Multi-centric database proved the generalization of the CNN model on different institutions across different continents. • CNN models can be used to aid on the diagnosis and follow-up of patients with prostate cancer.

Inteligencia artificial en las enfermedades respiratorias / Artificial Intelligence in Respiratory Diseases

No disponible

The use of µCT and fractal dimension for fracture prediction in osteoporotic individuals.

Osteoporosis (OP) is a widespread condition with commonly associated fracture sites at the hip, vertebra and wrist. This study examines the effects of age and osteoporosis on bone quality by comparing the efficacy of using parameters which indicate bone quality (both traditional clinical parameters such as bone mineral density (BMD), as well as apparent Young's modulus determined by finite element analysis, among others) to predict fracture. Non-fracture samples were collected from the femoral heads of 83 donors (44 males, 39 females), and fracture samples were obtained from the femoral heads of 17 donors (female). Microarchitectural parameters (Bone Volume/Total Volume [BV/TV], Bone Surface/Bone Volume [BS/BV], Tissue Mineral Density [TMD, etc.]) were measured from µCT of each sample as well as 2D and 3D fractal dimension (D2D and D3D respectively). A cube was cropped from µCT images and an isotropic hexahedral element was assigned to each voxel. Finite element analysis was used to calculate the Young's modulus for each sample. Overall, values for microarchitectural characteristics, fractal dimension measurements and Young's Modulus were consistent with values within literature. Significant correlations are observed between age and BV/TV for non-fracture males and females, as well as between age and volumetric BMD (vBMD) for the same groups. Significant differences are present between age-matched non-fracture and fracture females for BV/TV, BS/BV, vBMD, TMD, D2D, D3D, (p < 0.01 for all). Properties which are not age dependent are significantly different between age-matched non-fracture and fracture specimens, indicating OP is a disease, and not just an accelerated aging process.

Automated vertebrae localization and identification by decision forests and image-based refinement on real-world CT data.

PURPOSE: Development of a fully automatic algorithm for the automatic localization and identification of vertebral bodies in computed tomography (CT). MATERIALS AND METHODS: This algorithm was developed using a dataset based on real-world data of 232 thoraco-abdominopelvic CT scans retrospectively collected. In order to achieve an accurate solution, a two-stage automated method was developed: decision forests for a rough prediction of vertebral bodies position, and morphological image processing techniques to refine the previous detection by locating the position of the spinal canal. RESULTS: The mean distance error between the predicted vertebrae centroid position and truth was 13.7 mm. The identification rate was 79.6% on the thoracic region and of 74.8% on the lumbar segment. CONCLUSION: The algorithm provides a new method to detect and identify vertebral bodies from arbitrary field-of-view body CT scans.

Parametric finite element model of medial patellofemoral ligament reconstruction model development and clinical validation.

BACKGROUND: Currently, there is uncertainty regarding the long-term outcome of medial patellofemoral ligament reconstructions (MPFLr). Our objectives were: (1) to develop a parametric model of the patellofemoral joint (PFJ) enabling us to simulate different surgical techniques for MPFLr; (2) to determine the negative effects on the PFJ associated with each technique, which could be related to long-term deterioration of the PFJ. METHODS: A finite element model of the PFJ was created based on CT data from 24 knees with chronic lateral patellar instability. Patella contact pressure and maximum MPFL-graft stress at five angles of knee flexion (0, 30, 60, 90 and 120°) were analysed in three types of MPFLr: anatomic, non-anatomic with physiometric behaviour, and non-anatomic with non-physiometric behaviour. RESULTS: An increase in patella contact pressure was observed at 0 and 30° of knee flexion after both anatomic and non-anatomic MPFLr with physiometric behaviour. In both reconstructions, the ligament was tense between 0 and 30° of knee flexion, but at 60, 90 and 120°, it had no tension. In the third reconstruction, the behaviour was completely the opposite. CONCLUSION: A parametric model of the PFJ enables us to evaluate different types of MPFLr throughout the full range of motion of the knee, regarding the effect on the patellofemoral contact pressure, as well as the kinematic behaviour of the MPFL-graft and the maximum MPFL-graft stress.

Allopurinol partially prevents disuse muscle atrophy in mice and humans.

Disuse muscle wasting will likely affect everyone in his or her lifetime in response to pathologies such as joint immobilization, inactivity or bed rest. There are no good therapies to treat it. We previously found that allopurinol, a drug widely used to treat gout, protects muscle damage after exhaustive exercise and results in functional gains in old individuals. Thus, we decided to test its effect in the prevention of soleus muscle atrophy after two weeks of hindlimb unloading in mice, and lower leg immobilization following ankle sprain in humans (EudraCT: 2011-003541-17). Our results show that allopurinol partially protects against muscle atrophy in both mice and humans. The protective effect of allopurinol is similar to that of resistance exercise which is the best-known way to prevent muscle mass loss in disuse human models. We report that allopurinol protects against the loss of muscle mass by inhibiting the expression of ubiquitin ligases. Our results suggest that the ubiquitin-proteasome pathway is an appropriate therapeutic target to inhibit muscle wasting and emphasizes the role of allopurinol as a non-hormonal intervention to treat disuse muscle atrophy.

Development of imaging biomarkers and generation of big data.

Several image processing algorithms have emerged to cover unmet clinical needs but their application to radiological routine with a clear clinical impact is still not straightforward. Moving from local to big infrastructures, such as Medical Imaging Biobanks (millions of studies), or even more, Federations of Medical Imaging Biobanks (in some cases totaling to hundreds of millions of studies) require the integration of automated pipelines for fast analysis of pooled data to extract clinically relevant conclusions, not uniquely linked to medical imaging, but in combination to other information such as genetic profiling. A general strategy for the development of imaging biomarkers and their integration in the cloud for the quantitative management and exploitation in large databases is herein presented. The proposed platform has been successfully launched and is being validated nowadays among the early adopters' community of radiologists, clinicians, and medical imaging researchers.