Search

The large-scale availability of healthcare data provides significant opportunities for development of advanced Clinical Decision Support Systems that can enhance patient care. One such essential application is automated ICD-9 diagnosis group prediction, useful for a variety of healthcare delivery related tasks including documenting, billing and insurance claims. Past attempts considered patients’ multivariate lab events data and clinical text notes independently. To the best of our knowledge, ours is the first attempt to investigate the efficacy of integration of both these aspects for this task. Experiments on MIMIC-III dataset showed promising results.

Retrieving relevant answers from heterogeneous data formats, for given for questions, is a challenging problem. The process of pinpointing relevant information suitable to answer a question is further compounded in large document collections containing documents of substantial length. This paper presents the models designed as part of our submission to the DialDoc21 Shared Task (Document-grounded Dialogue and Conversational Question Answering) for span prediction in question answering. The proposed models leverage the superior predictive power of pretrained transformer models like RoBERTa, ALBERT and ELECTRA, to identify the most relevant information in an associated passage for the next agent turn. To further enhance the performance, the models were fine-tuned on different span selection based question answering datasets like SQuAD2.0 and Natural Questions (NQ) corpus. We also explored ensemble techniques for combining multiple models to achieve enhanced performance for the task. Our team SB_NITK ranked 6th on the leaderboard for the Knowledge Identification task, and our best ensemble model achieved an Exact score of 58.58 and an F1 score of 73.39.

Diabetic retinopathy (DR), a chronic disease in which the retina is damaged due to small vessel damage caused by diabetes mellitus, is one of the leading causes of vision impairment in diabetic patients. Detection of the earliest clinical sign of the advent of DR is a critical requirement for intervention and effective treatment. Ophthalmologists are trained to identify DR, based on examining specific minute changes in the eye - microaneurysms, retinal haemorrhages, macular edema and changes in the retinal blood vessels. Segmentation of microaneurysms (MA) is a critical requirement for the early diagnosis of DR and has been the primary focus of the research community over the past few years. In this work, a systematic review of existing literature is carried out to examine the diagnostic use of automated MA detection and segmentation for early DR diagnosis. We mainly focus on existing early DR diagnosis techniques to understand their strengths and weaknesses. Though early diagnosis is performed using colour fundus photography, fluorescein angiography or optical coherence tomography angiography images, our study is limited to colour fundus based techniques. The early DR diagnosis methodologies reviewed in this article can be broadly classified into classical image processing, conventional machine learning (ML), and deep learning (DL) based techniques. Though significant progress has been achieved in these three classes of early DR diagnosis, several challenges and gaps still exist, underscoring a considerable scope for the development of fully automated, user-friendly early DR diagnosis and grading systems. We discuss in detail the challenges that need to be addressed in designing such effective, efficient, and robust algorithms for early DR diagnosis systems and also the ample scope for future research in this area.

The fused-deposition modeling (FDM) process is carried out at an elevated temperature, preventing the addition of biological factors, drugs, bioactive compounds, etc, during fabrication. To overcome this disadvantage, a 3D interlinked porous polylactic acid (PLA) scaffold was fabricated by FDM, followed by the embedding of a polycaprolactone (PCL) scaffold into the pores of the PLA at room temperature, yielding a PLA-PCL scaffold. In addition, PLA-PCL scaffolds with nanohydroxyapatite (PLA-PCL-nHAP) and multiwalled carbon nanotubes (PLA-PCL-MWCNT) were also fabricated. Here, the FDM-fabricated PLA scaffold functions as the structural component, whereas the embedded PCL scaffold acts as the functional component, which provides a the ability to functionalize the scaffolds with the desired chemical or biological materials. The embedding process is straightforward, cost effective, and does not require sophistication. A mechanical characterization of the scaffolds suggests that the Young’s modulus of the PLA-PCL scaffold (16.02 MPa) was higher than that of the FDM-fabricated PLA (9.98 MPa) scaffold, by virtue of embedded PCL matrix. In addition, finite element analysis showed that the von Mises stress on a mandible with scaffolds was 4.04 MPa, whereas for a mandible with a defect, it was 6.7 MPa, confirming the stress distribution efficiency and mechanical stability of these scaffolds. Furthermore, field emission-scanning electron microscope analysis implied the presence of interlinked porous structures with pore diameters of 50 µm to 300 µm. X-ray diffraction results revealed an increased crystallinity (%) in the embedded models (PLA-PCL, PLA-PCL-nHAP and PLA-PCL-MWCNT), compared to a PLA printed scaffold. Additionally, Raman analysis revealed that the embedding process did not cause chemical alterations in the polymeric chains. In vitro analysis with human osteoblasts demonstrated the osteoconductive nature of the scaffold, which supported mineralization. In brief, the advantage of our model is that it helps to overcome the difficulties of manufacturing a filament with the desired additives for FDM, and offers the ability to incorporate the desired concentrations of heat-labile bioactive molecules during the embedding process at ambient temperatures.

Immune checkpoint inhibitors (ICIs) blocking programmed death receptor 1 or its ligand (anti-PD-1/PD-L1) are a burgeoning class of promising cancer treatments. However, not all patients respond to these treatments. Although studies have suggested potential biomarkers to predict patients likely to respond to treatment, no consensus biomarker has been identified. Here, 95 peer-reviewed studies representing 18,978 subjects across 15 cancer types were meta-analyzed to determine biomarkers that best predicted ICI response within and across cancers. Performance was assessed using the sensitivity, specificity, area under the receiver operating characteristic curve, and others. Across all cancers, multimodal biomarkers and tumor mutational burden discriminated ICI response better than PD-L1 immunohistochemistry assays (IHC) (P=.04) with sensitivities of 0.57 and 0.70 and specificities of 0.76 and 0.53, respectively. Marginal improvements were also observed for nascent microbiome biomarkers compared to IHC (P=.06). Current ICI biomarker performances indicate that additional research is needed to develop highly accurate and precise biomarkers for widespread clinical adoption.

Behavioral cues play a significant part in human communication and cognitive perception. In most professional domains, employee recruitment policies are framed such that both professional skills and personality traits are adequately assessed. Hiring interviews are structured to evaluate expansively a potential employee's suitability for the position - their professional qualifications, interpersonal skills, ability to perform in critical and stressful situations, in the presence of time and resource constraints, etc. Therefore, candidates need to be aware of their positive and negative attributes and be mindful of behavioral cues that might have adverse effects on their success. We propose a multimodal analytical framework that analyzes the candidate in an interview scenario and provides feedback for predefined labels such as engagement, speaking rate, eye contact, etc. We perform a comprehensive analysis that includes the interviewee's facial expressions, speech, and prosodic information, using the video, audio, and text transcripts obtained from the recorded interview. We use these multimodal data sources to construct a composite representation, which is used for training machine learning classifiers to predict the class labels. Such analysis is then used to provide constructive feedback to the interviewee for their behavioral cues and body language. Experimental validation showed that the proposed methodology achieved promising results., 9 pages, ACL 2020

Object detection has been at the forefront for higher level vision tasks such as scene understanding and contextual reasoning. Therefore, solving object detection for a large number of visual categories is paramount. Zero-Shot Object Detection (ZSD) – where training data is not available for some of the target classes – provides semantic scalability to object detection and reduces dependence on large amount of annotations, thus enabling a large number of applications in real-life scenarios. In this paper, we propose a novel multi-space approach to solve ZSD where we combine predictions obtained in two different search spaces. We learn the projection of visual features of proposals to the semantic embedding space and class labels in the semantic embedding space to visual space. We predict similarity scores in the individual spaces and combine them. We present promising results on two datasets, PASCAL VOC and MS COCO. We further discuss the problem of hubness and show that our approach alleviates hubness with a performance superior to previously proposed methods.

The detailed physiological perspectives captured by medical imaging provides actionable insights to doctors to manage comprehensive care of patients. However, the quality of such diagnostic image modalities is often affected by mismanagement of the image capturing process by poorly trained technicians and older/poorly maintained imaging equipment. Further, a patient is often subjected to scanning at different orientations to capture the frontal, lateral and sagittal views of the affected areas. Due to the large volume of diagnostic scans performed at a modern hospital, adequate documentation of such additional perspectives is mostly overlooked, which is also an essential key element of quality diagnostic systems and predictive analytics systems. Another crucial challenge affecting effective medical image data management is that the diagnostic scans are essentially stored as unstructured data, lacking a well-defined processing methodology for enabling intelligent image data management for supporting applications like similar patient retrieval , automated disease prediction etc. One solution is to incorporate automated diagnostic image descriptions of the observation/findings by leveraging computer vision and natural language processing. In this work, we present multi-task neural models capable of addressing these critical challenges. We propose ESRGAN, an image enhancement technique for improving the quality and visualization of medical chest x-ray images, thereby substantially improving the potential for accurate diagnosis, automatic detection and region-of-interest segmentation. We also propose a CNN-based model called ViewNet for predicting the view orientation of the x-ray image and generating a medical report using Xception net, thus facilitating a robust medical image management system for intelligent diagnosis applications. Experimental results are demonstrated using standard metrics like BRISQUE, PIQE and BLEU scores, indicating that the proposed models achieved excellent performance. Further, the proposed deep learning approaches enable diagnosis in a lesser time and their hybrid architecture shows significant potential for supporting many intelligent diagnosis applications.