Document Layout Analysis (DLA) is a segmentation process that decomposes a scanned document image into its blocks of interest and classifies them. DLA is essential in a large number of applications, such as Information Retrieval, Machine Translation, Optical Character Recognition (OCR) systems, and structured data extraction from documents. However, identification of document blocks in DLA is challenging due to variations of block locations, inter-and intra-class variability, and background noises. In this paper, we propose a novel texture-based convolutional neural network for document layout analysis, called DoT-Net. DoT-Net is a multiclass classifier that can effectively identify document component blocks such as text, image, table, mathematical expression, and line-diagram, whereas most related methods have focused on the text vs. non-text block classification problem. DoT-Net can capture textural variations among the multiclass regions of documents. Our proposed method DoT-Net achieved promising results outperforming state-of-the-art document layout classifiers on accuracy, F1 score, and AUC. The open-source code of DoT-Net is available at https://github.com/datax-lab/DoTNet.

Automatic histopathological Whole Slide Image (WSI) analysis for cancer classification has been highlighted along with the advancements in microscopic imaging techniques, since manual examination and diagnosis with WSIs are time- and cost-consuming. Recently, deep convolutional neural networks have succeeded in histopathological image analysis. However, despite the success of the development, there are still opportunities for further enhancements. In this paper, we propose a novel cancer texture-based deep neural network (CAT-Net) that learns scalable morphological features from histopathological WSIs. The innovation of CAT-Net is twofold: (1) capturing invariant spatial patterns by dilated convolutional layers and (2) improving predictive performance while reducing model complexity. Moreover, CAT-Net can provide discriminative morphological (texture) patterns formed on cancerous regions of histopathological images comparing to normal regions. We elucidated how our proposed method, CAT-Net, captures morphological patterns of interest in hierarchical levels in the model. The proposed method out-performed the current state-of-the-art benchmark methods on accuracy, precision, recall, and F1 score.

Industries can improve their business efficiency by analyzing and extracting relevant knowledge from large numbers of documents. Knowledge extraction manually from large volume of documents is labor intensive, unscalable and challenging. Consequently there have been a number of attempts to develop intelligent systems to automatically extract relevant knowledge from OCR documents. Moreover, the automatic system can improve the capability of search engine by providing application-specific domain knowledge. However, extracting the efficient information from OCR documents is challenging due to highly unstructured format [1, 11, 18, 26]. In this paper, we propose an efficient framework for a knowledge extraction system that takes keywords based queries and automatically extracts their most relevant knowledge from OCR documents by using text mining techniques. The framework can provide relevance ranking of knowledge to a given query. We tested the proposed framework on corpus of documents at GE Power where document consists of more than hundred pages in PDF.

Identifying strategies to enhance patient engagement and to control healthcare costs promotes a responsive and efficient healthcare system. The aim of this study is to predict healthcare cost savings associated with delivering telehealth advance care planning (ACP) to patients living with dementia. Two Geriatrics Workforce Enhancement Programs delivered training to primary care providers on using telehealth to provide ACP. Using electronic health records data from 6344 dual-eligible Medicare/Medicaid patients receiving telehealth primary care from trained providers in an urban safety net system, persons living with dementia (n = 401) were identified by extracting ICD-10 codes. The primary outcome was the estimated hospitalization-associated cost, with a key independent variable of ACP billing status. Multiple linear regressions and machine learning techniques estimated the impact of telehealth ACP on hospitalization-associated costs with a differential analysis by race. Compared to non-Hispanic Whites, hospitalization costs among Hispanic elders were higher by USD 14,232.40. Costs for non-English speakers or those having increased comorbidities were higher by USD 27,346.60 and USD 26,072.70, respectively. Overall, receiving ACP was associated with lower costs of USD 23,928.84. Dementia patients seen by primary care providers in a system receiving training to offer ACP via telehealth realized significant cost savings, with marked differences among those of non-White racial backgrounds.

Cholesterol is a key modulator of membrane protein structure and function, yet predicting cholesterol-binding sites remains challenging due to its non-druglike physicochemical properties. Here, we curated more than 800 high-resolution transmembrane protein structures containing cholesterol, and developed an interpretable, atom-based graph neural network, called CholBindNet. A positive-unlabeled (PU) training strategy was employed to address the scarcity of negative samples due to the promiscuous nature of cholesterol binding. We show that CholBindNet substantially outperforms existing machine learning models trained on general ligand-binding datasets, including AlphaFold3, P2Rank, and DiffDock. The performance and generalizability of the model on unseen membrane proteins were further demonstrated by rapidly assessing cholesterol-binding sites in the PIEZO2 ion channel against all-atom molecular dynamics (MD) simulations conducted on Anton3 supercomputer. Additionally, strong model interpretability was achieved for CholBindNet through atom-level feature encoding, Grad-CAM visualization, and attention-based scoring analysis. Overall, CholBindNet provides an efficient and scalable approach for classifying and ranking cholesterol-binding sites on membrane proteins, achieving performance comparable to computationally expensive MD simulations while offering rich biophysical insights into the atomic-level spatial patterns beyond amino-acid sequence. This work lays the foundation for future deep-learning models targeting membrane protein drug-binding sites and cholesterol-modulated therapeutics.