Prostate cancer segmentation from MRI by a multistream fusion encoder.

BACKGROUND: Targeted prostate biopsy guided by multiparametric magnetic resonance imaging (mpMRI) detects more clinically significant lesions than conventional systemic biopsy. Lesion segmentation is required for planning MRI-targeted biopsies. The requirement for integrating image features available in T2-weighted and diffusion-weighted images poses a challenge in prostate lesion segmentation from mpMRI. PURPOSE: A flexible and efficient multistream fusion encoder is proposed in this work to facilitate the multiscale fusion of features from multiple imaging streams. A patch-based loss function is introduced to improve the accuracy in segmenting small lesions. METHODS: The proposed multistream encoder fuses features extracted in the three imaging streams at each layer of the network, thereby allowing improved feature maps to propagate downstream and benefit segmentation performance. The fusion is achieved through a spatial attention map generated by optimally weighting the contribution of the convolution outputs from each stream. This design provides flexibility for the network to highlight image modalities according to their relative influence on the segmentation performance. The encoder also performs multiscale integration by highlighting the input feature maps (low-level features) with the spatial attention maps generated from convolution outputs (high-level features). The Dice similarity coefficient (DSC), serving as a cost function, is less sensitive to incorrect segmentation for small lesions. We address this issue by introducing a patch-based loss function that provides an average of the DSCs obtained from local image patches. This local average DSC is equally sensitive to large and small lesions, as the patch-based DSCs associated with small and large lesions have equal weights in this average DSC. RESULTS: The framework was evaluated in 931 sets of images acquired in several clinical studies at two centers in Hong Kong and the United Kingdom. In particular, the training, validation, and test sets contain 615, 144, and 172 sets of images, respectively. The proposed framework outperformed single-stream networks and three recently proposed multistream networks, attaining F1 scores of 82.2 and 87.6% in the lesion and patient levels, respectively. The average inference time for an axial image was 11.8 ms. CONCLUSION: The accuracy and efficiency afforded by the proposed framework would accelerate the MRI interpretation workflow of MRI-targeted biopsy and focal therapies.

Multimodal Image Classification by Multiview Latent Pattern Extraction, Selection, and Correlation.

The large amount of data available in the modern big data era opens new opportunities to expand our knowledge by integrating information from heterogeneous sources. Multiview learning has recently achieved tremendous success in deriving complementary information from multiple data modalities. This article proposes a framework called multiview latent space projection (MVLSP) to integrate features extracted from multiple sources in a discriminative way to facilitate binary and multiclass classifications. Our approach is associated with three innovations. First, most existing multiview learning algorithms promote pairwise consistency between two views and do not have a natural extension to applications with more than two views. MVLSP finds optimum mappings from a common latent space to match the feature space in each of the views. As the matching is performed on a view-by-view basis, the framework can be readily extended to multiview applications. Second, feature selection in the common latent space can be readily achieved by adding a class view, which matches the latent space representations of training samples with their corresponding labels. Then, high-order view correlations are extracted by considering feature-label correlations. Third, a technique is proposed to optimize the integration of different latent patterns based on their correlations. The experimental results on the prostate image dataset demonstrate the effectiveness of the proposed method.

Thickness-to-Height Ratio of Intravesical Prostatic Protrusion Predicts the Clinical Outcome and Morbidity of Prostatic Artery Embolization for Benign Prostatic Hyperplasia.

PURPOSE: To evaluate the presence of intravesical prostatic protrusion (IPP) and its thickness-to-height (T/H) ratio as a predictor for the clinical outcome and morbidity of prostatic artery embolization (PAE) for benign prostatic hyperplasia. MATERIALS AND METHODS: This was a prospective, single-center, institutional review board-approved study from June 2015 to December 2018 of 82 consecutive patients (age, 53-79 years; median, 66 years) with International Prostate Symptom Score (IPSS) ≥15 and quality-of-life (QOL) score ≥3. The presence of IPP and its T/H ratio were assessed on baseline magnetic resonance imaging for their correlation with the clinical outcomes of suboptimal IPSS (IPSS ≥10) and suboptimal QOL (QOL ≥3) up to 12 months after PAE and the occurrence of post-procedure complications (≤30 days), which caused a certain degree of urinary outflow obstruction. The chi-squared test was used for analysis. RESULTS: IPP was present in 57 of 82 patients (69.5%). The presence of IPP correlated with the occurrence of post-procedure complications (P = .009) but not with suboptimal IPSS at 12 months (P = .758). IPP with a T/H ratio ≤1.3 correlated with suboptimal IPSS at 12 months (P = .025) and suboptimal QOL at 6 months (P = .025) and 12 months (P = .008), as well as with the occurrence of post-procedure complications (P = .009). CONCLUSIONS: IPP with a T/H ratio ≤1.3 predicted the occurrence of post-procedure complications with urinary obstruction. A T/H ratio ≤1.3 but not the presence of IPP alone predicted the clinical outcome up to 12 months after PAE.

Shear wave elastography of thyroid nodules in routine clinical practice: preliminary observations and utility for detecting malignancy.

OBJECTIVE: To evaluate real-time shear wave ultrasound elastography (SWE) for characterizing focal thyroid lesions in routine clinical practice. METHODS: Seventy-four patients with 81 focal thyroid lesions undergoing conventional US with needle cytology also underwent SWE. Absolute and relative SWE stiffness measurements on colour-coded elastograms were correlated with cytology and their discriminatory performances assessed. RESULTS: Seventeen nodules were malignant (13 papillary, 4 other cancers), 45 benign (43 hyperplastic nodules, 2 focal thyroiditis), 5 indeterminate ("follicular lesions"), and 5 had inadequate cytology. SWE results were higher in malignant than benign nodules (P values 0.02-0.05) although their discriminatory performances were mediocre (AUCs 0.58-0.74). The most accurate SWE cut-off, 34.5 kPa for a 2-mm region of interest, achieved 76.9 % sensitivity and 71.1 % specificity for discriminating papillary cancer from benign nodules. No thresholds produced high sensitivity without lowering specificity appreciably, and vice versa. Nodule size correlated with SWE for benign nodules (P < 0.01). Intranodular cystic change or calcification did not influence SWE. Qualitatively, elastographic artefacts and foci lacking colour elasticity signal occurred in some solid nodules. CONCLUSION: Although malignant nodules are generally stiffer than benign nodules, the precision results do not suggest a definitive role for SWE, at present, in identifying or excluding thyroid malignancy. KEY POINTS : • Shear wave ultrasound elastography (SWE) offers new insight into thyroid disease. • Papillary cancers have higher SWE indices (equating to higher stiffness) than benign nodules. • SWE appears limited in terms of identifying or excluding thyroid malignancy accurately. • Vertically aligned elastographic artefacts can occur in thyroid SWE. • Areas lacking SWE colour signal can occur in some solid thyroid nodules.

A pilot study evaluating real-time shear wave ultrasound elastography of miscellaneous non-nodal neck masses in a routine head and neck ultrasound clinic.

A pilot study was performed to evaluate shear wave ultrasound elastography (SWE) for miscellaneous non-nodal/salivary/thyroid neck lesions. Forty-six lesions undergoing conventional sonography also underwent SWE. Elastic moduli from the stiffest areas in lesions were correlated with diagnosis. Forty lesions were benign (9 lipomas, 8 lymphatic/venous vascular malformations, 7 thyroglossal duct cysts, 4 branchial cleft cysts, 4 abscesses/phlegmons, 3 neurogenic tumors and 1 each of paraganglioma, sebaceous cyst, pseudotumor, hypertrophic scar, ranula) and 6 were malignant (1 malignant fibrous histiocytoma, 2 primary squamous cell carcinomas and 3 intramuscular metastases [2 squamous cell carcinomas, 1 malignant melanoma]).Median stiffness of malignant lesions (226.4 kPa, range 55.6 to 300.0) was higher than benign lesions (28.3 kPa, range 4.0 to 300.0) (p < 0.001). SWE cut-off with highest accuracy (174.4 kPa) achieved 83.3% sensitivity and 97.5% specificity, and the cut-off with 100% sensitivity (55.6 kPa) achieved 75% specificity. All malignant lesions were suspected on conventional sonography. The preliminary data indicate that SWE is feasible for miscellaneous neck lesions. SWE would not have altered management in terms of detecting undisclosed malignancies, although as a quantitative technique, it may increase the diagnostic confidence of less experienced operators performing head and neck ultrasound.

Shear wave elastography of focal salivary gland lesions: preliminary experience in a routine head and neck US clinic.

OBJECTIVES: To evaluate shear wave elastography (SWE) for focal lesions in major salivary glands. METHODS: Sixty lesions (49 parotid, 11 submandibular) undergoing routine ultrasound (grey scale and Doppler) also underwent SWE before US-guided needle aspiration for cytology. Quantitative indices of the shear elastic modulus (stiffness) were compared with cytological results. RESULTS: Fifty-five lesions were benign (21 pleomorphic adenomas, 18 Warthin's tumours; 16 others) and 5 malignant (2 mucoepidermoid carcinomas, 1 myoepithelial carcinoma, 1 B-cell lymphoma, 1 nodal metastasis). Shear modulus of benign lesions, median 18.3 kPa, overlapped appreciably with malignant lesions, median 13.5 kPa. However, 2 mucoepidermoid carcinomas had the highest stiffness values (81.9 kPa, 132.0 kPa). Stiffness of pleomorphic adenomas (median 22.5 kPa) was higher than Warthin's tumours (16.9 kPa) (P = 0.05 Mann-Whitney U-test). The standard deviation of stiffness values within a lesion, used as an indicator of spatial heterogeneity, was highest in mucoepidermoid cancers (median 44.2 kPa), followed by pleomorphic adenomas (median 12.4 kPa) and remaining lesions (medians 1.4-10.3 kPa). CONCLUSION: This study shows a degree of clustering of SWE indices according to pathology although it appears that SWE has suboptimal performance for ruling out malignancy, thus limiting its use in routine practice. KEY POINTS: • Shear wave elastography is a feasible technique for focal salivary gland lesions. • Elastographic artefacts aggravated by the regional anatomy may hinder this technique. • Elastographic indices vary according to pathology but there is appreciable overlap. • Overlapping indices for malignant and benign lesions limit its utility. • Pleomorphic adenomas have higher elasticity indices, i.e. are stiffer, than Warthin's tumours.

Shear wave elasticity imaging of cervical lymph nodes.

A pilot study of real-time shear wave ultrasound elastography (SWE) for cervical lymphadenopathy in routine clinical practice was conducted on 55 nodes undergoing conventional ultrasound (US) with US-guided needle aspiration for cytology. Elastic moduli of stiffest regions in nodes were measured on colour-coded elastograms, which were correlated with cytology. Malignant nodes (n = 31, 56.4%) were stiffer (median 25.0 kPa, range 6.9-278.9 kPa) than benign nodes (median 21.4 kPa, range 8.9-30.2 kPa) (p = 0.008, Mann Whitney U test). A cut-off of 30.2 kPa attained highest accuracy of 61.8%, corresponding to 41.9% sensitivity, 100% specificity and 0.77 area under the receiver operating characteristic curve. Qualitatively, elastograms of benign nodes were homogeneously soft; malignant nodes were homogeneously soft or markedly heterogeneous with some including regions lacking elasticity signal. SWE is feasible for neck nodes. It appears unsuitable for cancer screening but may detect a subset of malignant nodes. The cause of spatial heterogeneity of malignant nodes on SWE is yet to be established.

Real-time qualitative ultrasound elastography of cervical lymph nodes in routine clinical practice: interobserver agreement and correlation with malignancy.

To evaluate real-time qualitative ultrasound (US) elastography for cervical lymphadenopathy in routine clinical practice, 74 nodes (37 malignant, 37 benign) in 74 patients undergoing sonography underwent US elastography prior to fine needle aspiration for cytology. Dynamic cine loops of elasticity imaging displayed using a chromatic-scale were qualitatively scored by three independent observers for the proportion of stiff areas from ES1-4 (soft to stiff). There was fair to good interobserver agreement as indicated by weighted kappa (κ) statistic from 0.374 to 0.738. Median ES for benign and malignant nodes were 2 and 3 respectively. ES was higher in malignant nodes (p = 0.0003-0.0049, Mann Whitney U tests) although areas under receiver operating characteristic curves (0.68-0.74) indicated suboptimal discrimination. The optimal discriminatory cut-off, ES > 2, achieved only 62.2% sensitivity, 83.8% specificity and 73% accuracy for malignancy. Improvements in reliability and accuracy of real-time qualitative ultrasound elastography are required for it to be adopted into routine clinical practice.