In recent times, deep neural networks achieved outstanding predictive performance on various classification and pattern recognition tasks. However, many real-world prediction problems have ordinal response variables, and this ordering information is ignored by conventional classification losses such as the multi-category cross-entropy. Ordinal regression methods for deep neural networks address this. One such method is the CORAL method, which is based on an earlier binary label extension framework and achieves rank consistency among its output layer tasks by imposing a weight-sharing constraint. However, while earlier experiments showed that CORAL's rank consistency is beneficial for performance, the weight-sharing constraint could severely restrict the expressiveness of a deep neural network. In this paper, we propose an alternative method for rank-consistent ordinal regression that does not require a weight-sharing constraint in a neural network's fully connected output layer. We achieve this rank consistency by a novel training scheme using conditional training sets to obtain the unconditional rank probabilities through applying the chain rule for conditional probability distributions. Experiments on various datasets demonstrate the efficacy of the proposed method to utilize the ordinal target information, and the absence of the weight-sharing restriction improves the performance substantially compared to the CORAL reference approach.

Backgroundβ-Elemene, an effective anticancer component isolated from the Chinese herbal medicine Rhizoma Zedoariae, has been proved to have therapeutic potential against multiple cancers by extensive clinical trials and experimental research. However, its preventive role in cholangiocarcinoma (CCA) and the mechanisms of action of β-elemene on CCA need to be further investigated.MethodsA thioacetamide (TAA)-induced pre-CCA animal model was well-established, and a low dosage of β-elemene was intragastrically (i.g.) administered for 6 months. Livers were harvested and examined histologically by a deep-learning convolutional neural network (CNN). cDNA array was used to analyze the genetic changes of CCA cells following β-elemene treatment. Immunohistochemical methods were applied to detect β-elemene-targeted protein PCDH9 in CCA specimens, and its predictive role was analyzed. β-Elemene treatment at the cellular or animal level was performed to test the effect of this traditional Chinese medicine on CCA cells.ResultsIn the rat model of pre-CCA, the ratio of cholangiolar proliferation lesions was 0.98% ± 0.72% in the control group, significantly higher than that of the β-elemene (0. 47% ± 0.30%) groups (p = 0.0471). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis showed that the top 10 pathways affected by β-elemene treatment were associated with energy metabolism, and one was associated with the cell cycle. β-Elemene inactivated a number of oncogenes and restored the expression of multiple tumor suppressors. PCDH9 is a target of β-elemene and displays an important role in predicting tumor recurrence in CCA patients.ConclusionsThese findings proved that long-term use of β-elemene has the potential to interrupt the progression of CCA and improve the life quality of rats. Moreover, β-elemene exerted its anticancer potential partially by restoring the expression of PCDH9.

No abstract

Purpose: We aimed to construct machine learning (ML) radiomics models to predict response to lenvatinib monotherapy for unresectable hepatocellular carcinoma (HCC). Methods: Patients with HCC receiving lenvatinib monotherapy at three institutions were retrospectively identified and assigned to training and external validation cohorts. Tumor response after initiation of lenvatinib was evaluated. Radiomics features were extracted from contrast-enhanced computed tomography images. The K-means clustering algorithm was used to distinguish radiomics-based subtypes. Ten ML radiomics models were constructed and internally validated by 10-fold cross-validation. These models were subsequently verified in an external validation cohort. Results: A total of 109 patients were identified for analysis, namely, 74 in the training cohort and 35 in the external validation cohort. Thirty-two patients showed partial response, 33 showed stable disease and 44 showed progressive disease. The overall response rate (ORR) was 29.4% and the disease control rate (DCR) was 59.6%. A total of 224 radiomics features were extracted, and 25 significant features were identified for further analysis. Two distant radiomics-based subtypes were identified by K-means clustering, and subtype 1 was associated with a higher ORR and longer progression-free survival (PFS). Among the 10 ML algorithms, AutoGluon displayed the highest predictive performance (AUC=0.97), which was relatively stable in the validation cohort (AUC=0.93). Kaplan–Meier analysis showed that responders had a better overall survival (HR=0.21, 95% CI: 0.12-0.36, P<0.001) and PFS (HR=0.14, 95% CI: 0.09-0.22, P<0.001) than nonresponders. Conclusions: Valuable ML radiomics models were constructed, with favorable performance in predicting the response to lenvatinib monotherapy for unresectable HCC.

No abstract