We propose DeepChannel, a robust, data-efficient, and interpretable neural model for extractive document summarization. Given any document-summary pair, we estimate a salience score, which is modeled using an attention-based deep neural network, to represent the salience degree of the summary for yielding the document. We devise a contrastive training strategy to learn the salience estimation network, and then use the learned salience score as a guide and iteratively extract the most salient sentences from the document as our generated summary. In experiments, our model not only achieves state-of-the-art ROUGE scores on CNN/Daily Mail dataset, but also shows strong robustness in the out-of-domain test on DUC2007 test set. Moreover, our model reaches a ROUGE-1 F-1 score of 39.41 on CNN/Daily Mail test set with merely 1/100 training set, demonstrating a tremendous data efficiency.

Purpose: To propose a deep learning network with subregion partition for predicting metastatic origins and EGFR/HER2 status in patients with brain metastasis.Methods: We retrospectively enrolled 140 patients with clinico-pathologically confirmed brain metastasis originated from primary NSCLC (n = 60), breast cancer (BC, n = 60) and other tumor types (n = 20). All patients underwent contrast-enhanced brain MRI scans. The brain metastasis was subdivided into phenotypically consistent subregions using patient-level and population-level clustering. A residual network with a global average pooling layer (RN-GAP) was proposed to calculate deep learning-based features. Features from each subregion were selected with least absolute shrinkage and selection operator (LASSO) to build logistic regression models (LRs) for predicting primary tumor types (LR-NSCLC for the NSCLC origin and LR-BC for the BC origin), EGFR mutation status (LR-EGFR) and HER2 status (LR-HER2).Results: The brain metastasis can be partitioned into a marginal subregion (S1) and an inner subregion (S2) in the MRI image. The developed models showed good predictive performance in the training (AUCs, LR-NSCLC vs. LR-BC vs. LR-EGFR vs. LR-HER2, 0.860 vs. 0.909 vs. 0.850 vs. 0.900) and validation (AUCs, LR-NSCLC vs. LR-BC vs. LR-EGFR vs. LR-HER2, 0.819 vs. 0.872 vs. 0.750 vs. 0.830) set.Conclusion: Our proposed deep learning network with subregion partitions can accurately predict metastatic origins and EGFR/HER2 status of brain metastasis, and hence may have the potential to be non-invasive and preoperative new markers for guiding personalized treatment plans in patients with brain metastasis.

Zongmeng Zhang, Wengang Zhou, Jiaxin Shi, Houqiang Li. Findings of the Association for Computational Linguistics: ACL 2023. 2023.

Today's scene graph generation (SGG) task is still far from practical, mainly due to the severe training bias, e.g., collapsing diverse "human walk on / sit on / lay on beach" into "human on beach". Given such SGG, the down-stream tasks such as VQA can hardly infer better scene structures than merely a bag of objects. However, debiasing in SGG is not trivial because traditional debiasing methods cannot distinguish between the good and bad bias, e.g., good context prior (e.g., "person read book" rather than "eat") and bad long-tailed bias (e.g., "near" dominating "behind / in front of"). In this paper, we present a novel SGG framework based on causal inference but not the conventional likelihood. We first build a causal graph for SGG, and perform traditional biased training with the graph. Then, we propose to draw the counterfactual causality from the trained graph to infer the effect from the bad bias, which should be removed. In particular, we use Total Direct Effect (TDE) as the proposed final predicate score for unbiased SGG. Note that our framework is agnostic to any SGG model and thus can be widely applied in the community who seeks unbiased predictions. By using the proposed Scene Graph Diagnosis toolkit on the SGG benchmark Visual Genome and several prevailing models, we observed significant improvements over the previous state-of-the-art methods.

Background: Cholangiocarcinoma (CCA) are aggressive bile duct cancers with a poor prognosis for which there are only few established prognostic biomarkers and molecular targets available. The gene ITIH5, a known class II tumor suppressor gene (C2TSG), encodes a secreted protein of the extracellular matrix mediating tumor suppressive properties. Recently, it was surprisingly found that the ITIH5 protein is specifically upregulated in CCAs and that ITIH5 detection in blood could be an excellent liquid biopsy marker for indicating the presence of a CCA tumor in a patient. We therefore investigated whether patients with CCAs with abundant versus low ITIH5 protein expression also differ in their prognosis. Methods: To clarify this question, a large CCA cohort (n = 175) was examined using immunohistochemistry on a tissue microarray (TMA). Results: Abundant ITIH5 expression in CCA was associated with favorable survival, a low UICC stage and the absence of perineural invasion (PNI). Conclusions: ITIH5 has biomarker potential not only for the early detection of CCA from blood-based liquid biopsies but also as a prognostic tissue biomarker for risk stratification. Our results suggest that the upregulation of ITIH5 is particularly abundant in intrahepatic CCAs (iCCA). The mechanisms mediating the strong initial upregulation of ITIH5 during the oncogenic transformation of bile duct cells are still unclear.

Recent progress in variational inference has paid much attention to the flexibility of variational posteriors. One promising direction is to use implicit distributions, i.e., distributions without tractable densities as the variational posterior. However, existing methods on implicit posteriors still face challenges of noisy estimation and computational infeasibility when applied to models with high-dimensional latent variables. In this paper, we present a new approach named Kernel Implicit Variational Inference that addresses these challenges. As far as we know, for the first time implicit variational inference is successfully applied to Bayesian neural networks, which shows promising results on both regression and classification tasks.