Showing papers by "Pritee Khanna published in 2018"

An efficient algorithm for mass detection and shape analysis of different masses present in digital mammograms

Abstract: The present study introduces an efficient algorithm for automatic segmentation and detection of mass present in the mammograms. The problem of over and under-segmentation of low-contrast mammographic images has been solved by applying preprocessing on original mammograms. Subtraction operation performed between enhanced and enhanced inverted mammogram significantly highlights the suspicious mass region in mammograms. The segmentation accuracy of suspicious region has been improved by combining wavelet transform and fast fuzzy c-means clustering algorithm. The accuracy of mass segmentation has been quantified by means of Jaccard coefficients. Better sensitivity, specificity, accuracy, and area under the curve (AUC) are observed with support vector machine using radial basis kernel function. The proposed algorithm is validated on Mini-Mammographic Image Analysis Society (MIAS) and Digital Database for Screening Mammography (DDSM) datasets. Highest 91.76% sensitivity, 96.26% specificity, 95.46% accuracy, and 96.29% AUC on DDSM dataset and 94.63% sensitivity, 92.74% specificity, 92.02% accuracy, and 95.33% AUC on MIAS dataset are observed. Also, shape analysis of mass is performed by using moment invariant and Radon transform based features. The best results are obtained with Radon based features and achieved accuracies for round, oval, lobulated, and irregular shape of mass are 100%, 70%, 64%, and 96%, respectively.

Mesh-free based variational level set evolution for breast region segmentation and abnormality detection using mammograms.

Abstract: Automatic segmentation of abnormal region is a crucial task in computer aided detection system using mammograms. In this work an automatic abnormality detection algorithm using mammographic images is proposed. In the preprocessing step, partial differential equation based variational level set method is employed for breast region extraction. The evolution of the level set method is done by applying mesh-free based radial basis function. The limitation of mesh-based approach is removed by using mesh-free based RBF method. The evolution of variational level set function is also done by mesh-based finite difference method for comparison purpose. Unsharp masking and median filtering is employed for mammogram enhancement. Suspicious abnormal regions are segmented by applying fuzzy c-means clustering. Texture features are extracted from the segmented suspicious regions by computing local binary pattern and dominated rotated local binary pattern (DRLBP). Finally, suspicious regions are classified as normal or abnormal regions by means of support vector machine with linear, multilayer Perceptron, radial basis, and polynomial kernel function. The algorithm is validated on 322 sample mammograms of mammographic image analysis society (MIAS) and 500 mammograms from digital database for screening mammography (DDSM) datasets. Proficiency of the algorithm is quantified by using sensitivity, specificity, and accuracy. The highest sensitivity, specificity and accuracy of 93.96%, 95.01%, and 94.48%, respectively, are obtained on MIAS dataset using DRLBP feature with RBF kernel function. Whereas, the highest 92.31% sensitivity, 98.45% specificity, and 96.21% accuracy are achieved on DDSM dataset using DRLBP feature with RBF kernel function.

Classification of Food Images through Interactive Image Segmentation

Abstract: Food item segmentation in an image is a kind of fine-grained segmentation task, which is comparatively difficult than conventional image segmentation because intra-class variance is high and inter-class variance is low. So, an interactive food item segmentation algorithm using Random Forest is proposed in this work. The first step of the proposed algorithm is interactive food image segmentation, where food parts are extracted based on user inputs. It is observed that some of the segmented food parts may have some holes due to improper distribution of light. So, Boundary Detection & Filling and Gappy Principal Component Analysis methods are applied to restore the missing information in the second step. Local Binary Pattern and Non Redundant Local Binary Pattern are used for extracting features from the restored food parts, which are fed into support vector machine classifier for differentiating one food image from others. All the experiments have been performed on Food 101 database. A comparative study has also been done based on the three existing methods. The obtained results demonstrate that the proposed method outperforms the existing methods.

A statistical region selection and randomized volumetric features selection framework for early detection of Alzheimer's disease

Abstract: Identification of dominant imaging biomarkers is important for early detection of Alzheimer's disease (AD) and to improve diagnostic accuracy. This work proposes a novel automatic computer aided diagnosis (CAD) system working on region selection framework. Voxel based morphometry and tissue segmentation is performed to get gray matter (GM) images. These pre‐processed images are anatomized to get 116 regions of brain using a standard automated anatomical labeling atlas. The proposed region selection algorithm identifies the most relevant brain regions out of 116 regions to discriminate AD and healthy control (HC) subjects. Volumetric features (standard deviation, skewness, kurtosis, energy, and shannon entropy) are extracted and random feature selection is performed to get the most discriminating regions to classify AD from HC. Supervised classification algorithms are used to explore and validate the proposed methodology. Experimental results indicate that the performance of the proposed system competes well with the state‐of‐the‐art techniques.

Mesh-free approach for enhancement of mammograms

Abstract: Mammogram enhancement is a key step to detect breast cancer using digital mammogram. The present study investigates mesh-free radial basis function (RBF) collocation method to solve linear diffusion equation for image enhancement of mammograms. The proposed algorithm is compared with the mesh-based finite difference method as well as other existing enhancement methods such as unsharp masking, histogram equalisation, and contrast limited adaptive histogram equalisation. Specifically, figure-of-merits with emphasis on image contrast and computational time are assessed and compared with different image processing techniques. The proposed algorithm has been applied towards the enhancement of all 322 sample mammogram images of Mini-Mammographic Image Analysis Society and randomly selected 300 sample images from Digital Database for Screening Mammography databases. Finally, mean and standard deviation of contrast improvement index and peak signal-to-noise ratio have been estimated and presented. The outcome of this study indicates that the mesh-free-based RBF collocation method proves to be a computationally efficient and valuable contrast enhancement technique in an area of mammography.

A semi-supervised domain adaptation assembling approach for image classification

Abstract: Automatic annotation of images is one of the fundamental problems in computer vision applications. With the increasing amount of freely available images, it is quite possible that the training data used to learn a classifier has different distribution from the data which is used for testing. This results in degradation of the classifier performance and highlights the problem known as domain adaptation. Framework for domain adaptation typically requires a classification model which can utilize several classifiers by combining their results to get the desired accuracy. This work proposes depth-based and iterative depth-based fusion methods which are basically rank-based fusion methods and utilize rank of the predicted labels from different classifiers. Two frameworks are also proposed for domain adaptation. The first framework uses traditional machine learning algorithms, while the other works with metric learning as well as transfer learning algorithm. Motivated from ImageCLEF’s 2014 domain adaptation task, these frameworks with the proposed fusion methods are validated and verified by conducting experiments on the images from five domains having varied distributions. Bing, Caltech, ImageNet, and PASCAL are used as source domains and the target domain is SUN. Twelve object categories are chosen from these domains. The experimental results show the performance improvement not only over the baseline system, but also over the winner of the ImageCLEF’s 2014 domain adaptation challenge.

Optic Disc Segmentation using Vessel In-painting and Random Walk Algorithm

Abstract: Optic disc segmentation in fundus images is a fundamental step for the detection of retinal diseases like glaucoma. Glaucoma effects the parts of retina inside and around optic disc leading in manifestation of various structural abnormalities. The work proposed in this paper presents an efficient optic disc segmentation methodology using random walk algorithm. Random walk algorithm divides the image into foreground and background regions based on the initial seeds. The optic disc is segmented by using random walk with weights calculated on the color similarity and dissimilarity among neighborhood pixels. The proposed method is tested on fundus images of publicly available Drishti-GS1 database. The final performance is evaluated with respect to precision, sensitivity, specificity, F-score, jaccard, dice, and mean absolute distance measures and compared with other optic disc segmentation approaches presented in the literature.

Glioma identification from brain MRI using superpixels and FCM clustering

Abstract: This work presents a superpixel based computer aided diagnosis (CAD) system for brain tumor segmentation, classification, and identification of glioma tumors. It utilizes superpixel and fuzzy c-means clustering concept for tumor segmentation. At first, dataset images are preprocessed by anisotropic diffusion and dynamic stochastic resonance-based enhancement technique and further segmented through the proposed concept. The run length of centralized patterns are extracted from the segmented regions and classified with naive Bayes classifier. The performance of the system is examined on two brain magnetic resonance imaging datasets for segmentation and identification of glioma tumors. Accuracy for tumor detection is observed 99.89% on JMCD dataset and 100% on BRATS dataset. For glioma identification average accuracies are observed as 97.94% and 98.67% on JMCD and BRATS dataset, respectively. The robustness of the system is examined by 10-fold cross validation and statistical testing. Outcomes are also verified by domain experts in real time.

Classification of Brain MRIs Forming Superpixels

Abstract: The present work is focused on the classification of tumorous and non-tumorous images from brain magnetic resonance imaging (MRIs). Instead of pixelwise feature extraction, this work generates and uses superpixels for feature extraction. Superpixels reflect the group of pixels having specific properties. Run length of centralized patterns (RLCP) is extracted from each superpixel and concatenated to form a feature vector. Extracted features are classified using naive Bayes classifier. The experiments are performed on real datasets, one is collected from NSCB Medical College, Jabalpur (JMCD), India and the other is publically available Brain Tumor Segmentation Challenge (BRATS) dataset. The proposed computer-aided diagnosis (CAD) system based on superpixel segmentation outperforms other existing techniques and observes 99.84 and 100% accuracy on JMCD and BRATS, respectively. This well demonstrates that superpixel-based classification techniques can be developed for diagnosis of tumors on MR images used in the medical applications.