BackgroundClinical treatment decision making of bladder cancer (BCa) relies on the absence or presence of muscle invasion and tumor staging. Deep learning (DL) is a novel technique in image analysis, but its potential for evaluating the muscular invasiveness of bladder cancer remains unclear. The purpose of this study was to develop and validate a DL model based on computed tomography (CT) images for prediction of muscle-invasive status of BCa.MethodsA total of 441 BCa patients were retrospectively enrolled from two centers and were divided into development (n=183), tuning (n=110), internal validation (n=73) and external validation (n=75) cohorts. The model was built based on nephrographic phase images of preoperative CT urography. Receiver operating characteristic (ROC) curves were performed and the area under the ROC curve (AUC) for discrimination between muscle-invasive BCa and non-muscle-invasive BCa was calculated. The performance of the model was evaluated and compared with that of the subjective assessment by two radiologists.ResultsThe DL model exhibited relatively good performance in all cohorts [AUC: 0.861 in the internal validation cohort, 0.791 in the external validation cohort] and outperformed the two radiologists. The model yielded a sensitivity of 0.733, a specificity of 0.810 in the internal validation cohort and a sensitivity of 0.710 and a specificity of 0.773 in the external validation cohort.ConclusionThe proposed DL model based on CT images exhibited relatively good prediction ability of muscle-invasive status of BCa preoperatively, which may improve individual treatment of BCa.

No abstract

No abstract

Despite great strides made on fine-grained visual classification (FGVC), current methods are still heavily reliant on fully-supervised paradigms where ample expert labels are called for. Semi-supervised learning (SSL) techniques, acquiring knowledge from unlabeled data, provide a considerable means forward and have shown great promise for coarse-grained problems. However, exiting SSL paradigms mostly assume in-distribution (i.e., category-aligned) unlabeled data, which hinders their effectiveness when reproposed on FGVC. In this paper, we put forward a novel design specifically aimed at making out-of-distribution data work for semi-supervised FGVC, i.e., to "clue them in". We work off an important assumption that all fine-grained categories naturally follow a hierarchical structure (e.g., the phylogenetic tree of "Aves" that covers all bird species). It follows that, instead of operating on individual samples, we can instead predict sample relations within this tree structure as the optimization goal of SSL. Beyond this, we further introduced two strategies uniquely brought by these tree structures to achieve inter-sample consistency regularization and reliable pseudo-relation. Our experimental results reveal that (i) the proposed method yields good robustness against out-of-distribution data, and (ii) it can be equipped with prior arts, boosting their performance thus yielding state-of-the-art results. Code is available at https://github.com/PRIS-CV/RelMatch.