Informatics Publishing Limited

M. Akhil ¹, R. Aishwarya ¹, Vandhana Lal ¹, Shanthi Mahesh ²

Affiliations
1 Atria Institute of Technology, Anand Nagar, Bangalore - 560024, Karnataka, IN
2 Department of ISE, Atria Institute of Technology, Anand Nagar, Bangalore - 560024, Karnataka, IN

Abstract

Objective: Automated segmentation is an active research for medical images. Accuracy of automated segmentation methods plays a vital role during brain image analysis. Segmentation being an important area of research, determining its performance is also important. Gold Standard is required for comparison during segmentation evaluation. Method: The Gold Standard for segmentation of medical images is the manual drawing of region of interest. This manual tracing is performed by experts (radiologists). The deviation of segmentation when compared with the experts and the quality of segmentation are inversely proportional. Analysis: The quantitative methods indicate the performance of the segmentation methods when compared with Gold Standard. Evaluation metrics mostly fall into three categories: Area Based Evaluation method (Dice coefficient, Jaccard Coefficient, Relative Volume Difference, Volume Overlap error), Surface Evaluation type (Average Symmetric Surface Distance, Root Mean Square Symmetric Surface Distance, Scatter Plot) and Specificity, Sensitivity and Accuracy.

Keywords

Gold Standard, Segmentation, MRI, Manual Segmentation, Automated Segmentation, Evaluation Metrics.

Full Text

   

Suhas A. Bhyratae ¹, Shanthi Mahesh ², A. S. Harshith ¹, Harshitha C. Shekar ¹

Affiliations
1 Department of ISE, Atria Institute of Technology, ASKB Campus, 1st Main Road, AGS Colony, Anand Nagar, Bengaluru - 560024, IN
2 Information Science and Engineering, Atria Institute of Technology, ASKB Campus, 1st Main Road, AGS Colony, Anand Nagar, Bengaluru - 560024, Karnataka, IN

Abstract

Objectives: The objective of the proposing system is to classify Computed Tomographyimages into two classifications like cancernodes and without cancernodes for lung pictures using SVM and PSO classifiers. Methods/Analysis: Computed tomography (CT) cancerous lung images are used. Techniques employed include the usage of support vector machine and particle swarm optimization classifiers. Proposed techniques include histogram equalization, ROI masking using thresholding, wavelet transform, gray level co-occurrence matrix. Hence resulting in results in high accuracy, also yields better result in terms of the recognition accuracy. Findings: SVM shows higher accuracy rate of identifying the cancerous nodules than PSO for different iterations held. These comparison results between SVM and PSO bring out the efficiency of comparing between the two classifications algorithms. In the existing system 1.afusion of classification techniques was used to determine the lung nodes in the CT image and feature extraction based technique was implemented on the node. 2. As the second step, the features were computed based on its texture to distinguish the blood vessels. The proposed system is implemented to classify by combining of SVM and PSO. Hence leading to Low performance and accuracy in classification. With the results obtained in the proposed system it differs in various aspects like the lung nodes are well identified, better result obtained due to comparisons hence highlights the fact that proposed system had far more advantages than the existing system.

Keywords

Computed Tomography, Gene Selection, Particle Swam Optimization, Support Vector Machine.

Full Text

   

Neha Mangla ¹, Shanthi Mahesh ¹, M. Chhaya ¹, G. Vidyashree ¹, Vikas ¹

Affiliations
1 Atria Institute of Technology, Bangalore, IN

Abstract

Voluminous amount of structured, semistructured and unstructured data sets that have the potential to learn the relationship among data in the area of business is being collected rapidly; termed as big data. The storage of large chunks of data is difficult as even terabytes and petabytes of traditional data warehousing solutions is insufficient and exorbitant.
It is viable to store and process these ransom amount of data on Hadoop; which is a low cost, reliable, scalable and fault tolerant Java-based programming framework that supports the processing of large data sets in a distributed computing environment. Hadoop implements MapReduce programing model for storing and processing large data sets with a parallel, distributed algorithm on commodity hardware. Nevertheless, the programming model expects the developers to write bespoke programs that are less flexible, time consuming, hard to code; maintain and reuse. This challenging task of writing complex MapReduce codes was rationalized by making use of HiveQL.
Hive is the platform required to run HiveQL. Hive is built on top of Hadoop to query Big Data. Internally the Hive queries are converted into the corresponding MapReduce task.
In this paper, by making use of machine learning algorithm a movie rating prediction system is built based on MovieLens dataset.

Keywords

Big Data, HDFS, Hadoop, Hive, MapReduce, Linear Regression.

Full Text