The Modified Assessment Tool Based on Scapular Y-View for Global Fatty Infiltration in the Supraspinatus Muscle: Correlation, Predictive Performance, and Reliability Analyses.

BACKGROUND: The accurate evaluation of rotator cuff (RC) fatty degeneration after tears is critical for appropriate surgical decision making and prognosis. However, there is currently no reliable and practical tool to reflect the global fatty infiltration (Global-FI) throughout the 3-dimensional (3D) volumetric RC muscles. PURPOSE: (1) To determine the correlations between 2 modified assessment tools and the Global-FI and their predictive performances and reliabilities for Global-FI prediction, and (2) to compare these predictive parameters with those of the conventional tool using a single scapular Y-view slice. STUDY DESIGN: Cohort study (diagnosis); Level of evidence, 3. METHODS: A total of 49 patients with full-thickness RC tears scheduled to undergo arthroscopic repairs were included, and their surgical shoulders underwent 6-point Dixon magnetic resonance imaging preoperatively. The Global-FI was measured by calculating the 3D-volumetric fat fraction (FF) of the whole supraspinatus muscle through all acquired oblique sagittal slices. As a commonly used radiological landmark, the scapular Y-view was used to evaluate single-plane fatty infiltration (Y-FI) by calculating the FF in 1 slice, defined as the conventional assessment tool. Two modified assessment tools expand the analytic imaging by integrating the FFs from the scapular Y-view slice and its neighboring slices, which were calculated by averaging the FFs of these 3 slices (meanY3-FI) and accumulating local 3D-volumetric FFs from 3 slices (volY3-FI), respectively. The correlations between 3 assessment tools and the Global-FI were analyzed, and the predictive performance for Global-FI prediction using these tools was determined by goodness of fit and agreement. Moreover, the inter- and intraobserver reliabilities of these assessment tools were evaluated. Similar analyses were performed in the small-medium, large, or massive tear subgroups. RESULTS: The Y-FI was significantly higher than the Global-FI in all cases and tear size subgroups, while the 2 modified assessment tools (meanY3-FI and volY3-FI) did not significantly differ from the Global-FI. All assessment tools were significantly correlated with the Global-FI, but the meanY3-FI and volY3-FI showed stronger correlations than the Y-FI, which was also determined in different tear sizes. Moreover, the regression models of the meanY3-FI and volY3-FI showed superior goodness of fit to Y-FI in Global-FI prediction in all cases and subgroups, with larger coefficients of determination (R2) and smaller root mean square errors. The predicted Global-FI using the regression model of volY3-FI had the best agreement with the measured Global-FI, followed by the meanY3-FI, both showing smaller biases and standard deviation of the percentage difference between predicted- and measured Global-FI than the conventional Y-FI. In addition, the 2 modified assessment tools achieved better interobserver and intraobserver reliabilities than the conventional tool in all cases and subgroups. CONCLUSION: Two modified assessment tools (meanY3-FI and volY3-FI) were comparable with the Global-FI of the whole supraspinatus muscle, showing stronger correlations with the Global-FI and better predictive performances and reliabilities than the conventional tool (Y-FI). Moreover, the volY3-FI was slightly superior to meanY3-FI in the predictive performance and reliability.

Ranking-Preserving Cross-Source Learning for Image Retargeting Quality Assessment.

Image retargeting techniques adjust images into different sizes and have attracted much attention recently. Objective quality assessment (OQA) of image retargeting results is often desired to automatically select the best results. Existing OQA methods train a model using some benchmarks (e.g., RetargetMe), in which subjective scores evaluated by users are provided. Observing that it is challenging even for human subjects to give consistent scores for retargeting results of different source images (diff-source-results), in this paper we propose a learning-based OQA method that trains a General Regression Neural Network (GRNN) model based on relative scores-which preserve the ranking-of retargeting results of the same source image (same-source-results). In particular, we develop a novel training scheme with provable convergence that learns a common base scalar for same-source-results. With this source specific offset, our computed scores not only preserve the ranking of subjective scores for same-source-results, but also provide a reference to compare the diff-source-results. We train and evaluate our GRNN model using human preference data collected in RetargetMe. We further introduce a subjective benchmark to evaluate the generalizability of different OQA methods. Experimental results demonstrate that our method outperforms ten representative OQA methods in ranking prediction and has better generalizability to different datasets.