Showing papers by "Anupam Shukla published in 2020"

Detection of Pediatric Pneumonia from Chest X-Ray Images using CNN and Transfer Learning

Abstract: Pneumonia is one of the most fatal diseases caused in the lungs. The diagnosis involves a chest x-ray which is interpreted by a radiologist. Human assisted diagnosis has its own limitations like the availability of an expert, cost, etc and hence an automated method for the detection of pneumonia from x-rays is a necessity. In this research, neural network models were developed to detect pneumonia from the chest x-ray images. Four models namely a basic convolutional neural network (CNN), VGG16, VGG19, InceptionV3 were constructed using CNN and transfer learning methodologies. The models were then trained on a pediatric pneumonia dataset which comprised of 2992 pneumonia and 2972 normal chest xrays. The results were then tested using 854 pneumonia and 849 normal images, and an accuracy of over 97 percent was obtained from all models.

A solution for priority-based multi-robot path planning problem with obstacles using ant lion optimization

Abstract: This paper provides a Priority-based Optimized-Hybrid Asynchronous Centralized and Decentralized (POHACD) algorithm for multi-robot path planning (MRPP) issue, where multiple robots communicate wit...

Sign Language Recognition Using ResNet50 Deep Neural Network Architecture

Abstract: Communication is a barrier between the deaf-mute community and the rest of the society. Sign Language is used for communication among such people who cannot speak and listen. The automation of sign language recognition has gained researchers attention in the last few years. Many complex and costly hardware systems have already been developed to assist the purpose. However, we propose to use deep learning approach for automated sign language recognition. We devised a novel 2-level ResNet50 based Deep Neural Network Architecture to classify fingerspelled words. The dataset used is the standard American Sign Language Hand gesture dataset by [1]. The dataset was first augmented using various augmentation techniques. In our 2-level ResNet50 based approach the Level 1 model classifies the input image into one of the 4 sets. After an image is classified into one of the sets it is provided as an input to the corresponding second level model for predicting the actual class of the image. Our approach yields an accuracy of 99.03% on 12,048 test images.

Deep Reinforcement Learning for Single-Shot Diagnosis and Adaptation in Damaged Robots

Abstract: Robotics has proved to be an indispensable tool in many industrial as well as social applications, such as warehouse automation, manufacturing, disaster robotics, etc. In most of these scenarios, damage to the agent while accomplishing mission-critical tasks can result in failure. To enable robotic adaptation in such situations, the agent needs to adopt policies which are robust to a diverse set of damages and must do so with minimum computational complexity. We thus propose a damage aware control architecture which diagnoses the damage prior to gait selection while also incorporating domain randomization in the damage space for learning a robust policy. To implement damage awareness, we have used a Long Short Term Memory based supervised learning network which diagnoses the damage and predicts the type of damage. The main novelty of this approach is that only a single policy is trained to adapt against a wide variety of damages and the diagnosis is done in a single trial at the time of damage.

A comprehensive evaluation for the prediction of mortality in intensive care units with LSTM networks: patients with cardiovascular disease

Abstract: Intensive care units (ICUs) are responsible for generating a wealth of useful data in the form of electronic health records. We aimed to build a mortality prediction model on a Medical Information Mart for Intensive Care (MIMIC-III) database and to assess whether the use of deep learning techniques like long short-term memory (LSTM) can effectively utilize the temporal relations among clinical variables. The models were built on clinical variable dynamics of the first 48 h of ICU admission of 12,550 records from the MIMIC-III database. A total of 36 variables including 33 time series variables and three static variables were used for the prediction. We present the application of LSTM and LSTM attention (LSTM-AT) model for mortality prediction with such a large number of clinical variables dataset. For training and validation purpose, we have used International Classification of Diseases, 9th edition (ICD-9) codes for extracting the patients with cardiovascular disease, and infections and parasitic disease, respectively. The effectiveness of the LSTM model is achieved over non-recurrent baseline models like naive Bayes, logistic regression (LR), support vector machine and multilayer perceptron (MLP) by generating state of the art results (area under the curve [AUC], 0.852). Next, by providing attention at each time stamp, we developed a model, LSTM-AT, which exhibits even better performance (AUC, 0.876).

Deep-learning approaches for health care: Patients in intensive care

Abstract: Providing the appropriate care for patients admitted to intensive care units (ICUs) is becoming increasingly complex and difficult, thus creating a need for the use of deep learning models. Real-time predictions provided by such models can aid physicians in interpreting the vast amount of patient data generated in the ICU. A huge percentage of electronic health record (EHR) data is generated through ICUs. Data for this study were extracted from MIMIC-III (Medical Information Mart for Intensive Care), a time-series database of intensive care unit (ICU) patient data. Applications of deep learning and nature-inspired algorithms for health care have evolved over the last few years. In this research, different deep-learning algorithms, such as long short-term memory (LSTM) and long short-term memory-attention (LSTM-AT), are investigated, using discharge notes and clinical and laboratory recordings, for hospital readmission and mortality prediction tasks. We have used explicit approaches for both the applications. An AUC (area under curve) of 0.72 for the hospital readmission task utilizing the discharge notes and an AUC of 0.87 for mortality prediction utilizing the bedside recordings were achieved.