In recent years, several deep learning networks are proposed to segment 2D or 3D bio-medical images. However, in liver and lesion segmentation, the proportion of interested tissues and lesions are tiny when contrasting to the image background. That is, the objects to be segmented are highly imbalanced in terms of the frequency of occurrences. This makes existing deep learning networks prone to predict pixels of livers and lesions as background. To address this imbalance issue, several loss functions are proposed. Since no researches are having made a comparison among those proposed loss functions, we are curious about that which loss function is the best among them? At the same time, we also want to investigate whether the combination of several different loss functions is effective for liver and lesion segmentation. Firstly, we propose a novel deep learning network (cascade U-ResNets) to produce liver and lesion segmentation simultaneously. Then, we investigate the performance of 5 selected loss functions, WCE (Weighted Cross Entropy), DL (Dice Loss), WDL (Weighted Dice Loss), TL (Teverskry Loss), WTL (Weighted Teversky Loss), with our cascade U-ResNets. We further assemble all cascade U-ResNets trained with different loss functions together to segment livers and lesions jointly on the liver CT (Computed Tomography) volume. Experimental results on the LiTS dataset1 showed our ensemble model can achieve much better results than every individual model for liver segmentation.

Essays are a pivotal component of conventional exams; accurately, efficiently, and effectively grading them is a significant challenge for educators. Automated essay scoring (AES) is a complex task that utilizes computer technology to assist teachers in scoring. Traditional AES techniques only focus on shallow linguistic features based on the grading criteria, ignoring the influence of deep semantic features. The AES model based on deep neural networks (DNN) can eliminate the need for feature engineering and achieve better accuracy. In addition, the DNN-AES model combining different scales of essays has recently achieved excellent results. However, it has the following problems: (1) It mainly extracts sentence-scale features manually and cannot be fine-tuned for specific tasks. (2) It does not consider the shallow linguistic features that the DNN-AES cannot extract. (3) It does not contain the relevance between the essay and the corresponding prompt. To solve these problems, we propose an AES method based on multi-scale features. Specifically, we utilize Sentence-BERT (SBERT) to vectorize sentences and connect them to the DNN-AES model. Furthermore, the typical shallow linguistic features and prompt-related features are integrated into the distributed features of the essay. The experimental results show that the Quadratic Weighted Kappa of our proposed method on the Kaggle ASAP competition dataset reaches 79.3%, verifying the efficacy of the extended method in the AES task.

Medical data is an important part of modern medicine. However, with the rapid increase in the amount of data, it has become hard to use this data effectively. The development of machine learning, such as feature engineering, enables researchers to capture and extract valuable information from medical data. Many deep learning methods are conducted to handle various subtasks of EHR from the view of information extraction and representation learning. This survey designs a taxonomy to summarize and introduce the existing deep learning-based methods on EHR, which could be divided into four types (Information Extraction, Representation Learning, Medical Prediction and Privacy Protection). Furthermore, we summarize the most recognized EHR datasets, MIMIC, eICU, PCORnet, Open NHS, NCBI-disease and i2b2/n2c2 NLP Research Data Sets, and introduce the labeling scheme of these datasets. Furthermore, we provide an overview of deep learning models in various EHR applications. Finally, we conclude the challenges that EHR tasks face and identify avenues of future deep EHR research.

Multi-Label Continual Learning (MLCL) is a framework designed for class-incremental multi-label image recognition. However, MLCL faces two critical challenges: the construction of label relationships on past-missing and future-missing partial labels of training data, and the problem of catastrophic forgetting , which leads to poor generalization. To address these challenges, this study proposes an enhanced version of the Augmented Graph Convolutional Network (AGCN++), capable of constructing cross-task label relationships and mitigating catastrophic forgetting. First, an Augmented Correlation Matrix (ACM) is constructed across all observed classes, incorporating intra-task relationships derived from hard label statistics. Additionally, inter-task relationships are established by leveraging both hard and soft labels obtained from the data, as well as a constructed expert network. Next, a novel partial label encoder (PLE) is introduced for MLCL, enabling the extraction of dynamic class representations for each partial label image as graph nodes. This PLE also facilitates the generation of soft labels, which contribute to the creation of a more persuasive ACM and effectively mitigate forgetting. Lastly, a relationship-preserving constrainter is proposed to address the issue of forgetting label dependencies across old tasks. In the AGCN++, the label relationships topology can be augmented automatically, thereby generating efficient class representations. The effectiveness of the proposed method is evaluated using two multi-label image benchmarks. The experimental results demonstrate that the proposed approach is highly effective in the context of MLCL image recognition. It can establish compelling correlations across tasks, even in scenarios where the old task labels are missing.

As one of the essential pieces of evidence of crime scenes, footprint images cannot be ignored in the cracking of serial cases. Traditional footprint comparison and retrieval require much time and human resources, significantly affecting the progress of the case. With the rapid development of deep learning, the convolutional neural network has shown excellent performance in image recognition and retrieval. To meet the actual needs of public security footprint image retrieval, we explore the effect of convolution neural networks on footprint image retrieval and propose an ensemble deep neural network for image retrieval based on transfer learning. At the same time, based on edge computing technology, we developed a footprint acquisition system to collect footprint data. Experimental results on the footprint dataset we built show that our approach is useful and practical.