Papers by Gayatri Mirajkar
Gayatri Mirajkar, 2020
Investigation of blood smear is a significant symptomatic test utilized in the determination of a... more Investigation of blood smear is a significant symptomatic test utilized in the determination of a variety of sicknesses. The strategy for programmed conclusion of minute blood smear pictures by recognizing and isolating it to various classes of cells that are featured in the paper. Robotization of this cycle eventually limits the extent of potential illnesses saving a extensive measure of time. Recognizable proof of red platelets (RBCs) is completed/done by the framework utilizing unique procedures of picture handling activities like pre-handling, tasks for morphology, marking and extraction of highlights to ascertain shape and size of the RBCs. Morphological properties can give data in regards to state of the cell. By these activities and computations, RBCs are arranged. There are 2 phases in red cell arrangement process, first is the detachment of RBCs to typical and strange followed by strange cells grouping to three subclasses in view of the cell shape and construction. The point of this framework is to help pathologist by giving fast outcomes by examining the smear tests. The fix of these sicknesses is conceivable, when it is recognized at a previous stage.
G. Mirajkar, 2021
Detecting and classifying brain tumors is a critical and time-consuming task for medical professi... more Detecting and classifying brain tumors is a critical and time-consuming task for medical professionals. To expedite this process and ensure accuracy, we explored the application of advanced technology, specifically deep learning models, for medical image segmentation. Our focus was on identifying a robust model for brain tumor segmentation using a public MRI imaging dataset consisting of 3064 TI-weighted images from 233 patients with meningioma, glioma, and pituitary tumors. In our study, we meticulously converted and preprocessed the dataset before delving into the methodology. We implemented and trained well-established image segmentation deep learning models, including U-Net, Attention U-Net with various backbones, Deep Residual U-Net, ResUnet++, and Recurrent Residual U-Net. We varied the parameters based on our comprehensive review of the literature on human brain tumor classification and segmentation. the applied approaches, the recurrent residual U-Net utilizing the Adam optimizer achieved a Mean Intersection Over Union of 0.8665. This model outperformed other state-of-the-art deep learning models in terms of accuracy. Visual findings showcased remarkable results in brain tumor segmentation from MRI scans, highlighting the algorithm's potential to automatically extract brain cancers and assist physicians in serving humanity more effectively. The efficiency of this approach offers promising implications for expediting diagnosis and treatment planning in the realm of neuro-oncology.
Gayatri Mirajkar, 2021
To recognize deforestation utilizing Earth Observation (EO) information, generally utilized techn... more To recognize deforestation utilizing Earth Observation (EO) information, generally utilized techniques depend on the identification of fleeting changes in the EO estimations inside the deforested patches. In this paper, we present another mark of deforestation got from manufactured gap radar (SAR) pictures, which depends on a mathematical curio that seems when deforestation occurs, as a shadow at the boundary of the deforested area. The circumstances for the presence of these shadows are investigated, too as the strategies that can be utilized to take advantage of them to distinguish deforestation. The methodology includes two stages: (1) location of new shadows; (2) remaking of the deforested fix around the shadows. The data from Sentinel-1 of 2014 has opened up valuable open doors for a likely application of this methodology in huge scope applications. A deforestation identification technique in view of this approach was tried in a 600,000 ha site in Peru. A recognition accuracy of over 95% is gotten for tests bigger than 0.4 ha, and the strategy was found to perform better compared to the optical-based UMD-GLAD Alerts GLAD Forest Alert dataset both with regards to spatial and transient identification. Further work expected to take advantage of this methodology at functional levels is also considered.
Gayatri Mirajkar, 2021
Brain tumours are one of the most lethal types of cancer, and early detection is crucial for succ... more Brain tumours are one of the most lethal types of cancer, and early detection is crucial for successful treatment. Magnetic Resonance (MR) imaging is a widely used diagnostic tool for brain tumours. However, the interpretation of MR images can be challenging, and a trained radiologist is required for accurate diagnosis. In this study, we propose an automated system for the early detection of brain tumours using Deep Learning techniques. Our system uses a Convolutional Neural Network (CNN) to analyse MR images and classify them as either healthy or containing a tumour. The proposed system was trained on a large dataset of MR images and achieved high accuracy in tumour detection. The results demonstrate the potential of Deep Learning techniques for the early detection of brain tumours, which could lead to improved patient outcomes.
G. Mirajkar, 2022
SAR pictures have shown to be a useful technique for determining the age of forest stands. For th... more SAR pictures have shown to be a useful technique for determining the age of forest stands. For this aim, texture analysis methods have been frequently employed. We compared three texture analysis approaches in this study: Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP), and Gabor Filters. We collected texture characteristics from each approach using SAR photos of a wooded region in Canada. The retrieved characteristics were then statistically examined to see how efficient they were at determining forest stand age. Our findings demonstrate that GLCM and LBP outperformed Gabor Filters in determining forest stand age. This work adds to the expanding body of information on the use of SAR pictures and texture analysis methods for determining the age of forest stands.
G. Mirajkar, 2022
Deforestation is one of the major natural issues that are influencing this present reality. To sc... more Deforestation is one of the major natural issues that are influencing this present reality. To screen and identify deforestation, different remote detecting procedures have been created throughout the long term. In any case, the discovery of deforestation utilizing Engineered Gap Radar (SAR) information has been an area of dynamic examination lately. SAR information enjoys the benefit of having the option to enter mists and is accordingly more valuable in regions with weighty overcast cover. In this paper, we present a profound learning-based approach for the location of deforestation utilizing SAR information. Our methodology depends on Convolutional Brain Organizations (CNNs) and involves SAR information as info. We show that our methodology beats customary AI procedures for deforestation discovery.
Advances in computational sciences and technology, Dec 30, 2011
The optimal feature selection has been done using ICA and significant feature processing applied ... more The optimal feature selection has been done using ICA and significant feature processing applied to human brain MR images. Implementation in MR image segmentation has been studied significantly with kmeans and fuzzy c -means algorithm using MATLAB 2009a. The input to the kmeans algorithm and fuzzy c-means algorithm is the final feature image obtained by processing the outputs of Gabor wavelets and the number of desired classes in which the input image must be segmented. The result obtained from the clustering algorithm is the tumor region delineated from the unaffected brain tissue. Both ICA as well as significant feature processing gives the resultant feature image that contains the tumor and edema highlighted from the surrounding brain tissue. It has been observed that both methods are effective even in the presence of complex structures. ICA provides an effect that is similar to smoothing of the images resulting in absolute segmentation. The statistical method provides sharp segmentation results. Clustering, magnetic resonance image, independent component analysis, significant feature processing, kmeans, fuzzy c-means.
Advances in computational sciences and technology, Dec 30, 2011
In the magnetic resonance image segmentation the k-means algorithm and fuzzy c-means clustering a... more In the magnetic resonance image segmentation the k-means algorithm and fuzzy c-means clustering algorithm have been developed and implemented. Simulations have been done in MATLAB 2010b. Applying the expectationmaximization algorithm directly to the MR image of human brain does not provide satisfactory results in the segmentation of human brain tumor. The input to the kmeans algorithm is the final feature image obtained by processing the outputs of Gabor wavelets and the number of desired classes in which the input image must be segmented. The region with the highest intensity in the tumor has been labeled as the darkest in the kmeans algorithm. Kmeans clustering for features extracted from all three levels of approximation has been study thoroughly. Fuzzy c-means clustering is applied to the feature image obtained from combining the Gabor outputs. The clustering algorithm evaluates the six classes in the feature image. Within each class, the candidate pixels showing the greatest tendency of belonging to the particular class are highlighted with the darkest colors. In fuzzy c-means clustering, at every level of decomposition, the tumor is perfectly segmented out showing its presence as an object at all levels of approximation which is definite advantage over kmeans clustering algorithm.
Advances in computational sciences and technology, Dec 30, 2014
Skull stripping forms an important pre-processing step in neuroimaging analysis. In this paper, a... more Skull stripping forms an important pre-processing step in neuroimaging analysis. In this paper, a local Chan-Vese Expectation Maximization (LCV-EM) model is proposed for skull segmentation which uses both global image information and the local information obtained via the Expectation Maximization (EM) algorithm. The energy functional for the proposed model consists of three terms: the global term, the local EM term, and the regularization term. Since magnetic resonance (MR) images contain a lot of intensity in homogeneity, the use of the local EM term along with the global term allows the segmentation of the brain from the skull and the non-brain tissue, in spite of the partial volume effect prominent near the boundary of the skull. The LCV-EM model is applicable to both T1 and T2-weighted MR images. The proposed model has the advantage that it does not require any boundary function or stopping function to decide the true boundary of the skull. Also the model shows good performance in comparison with other methods for brain extraction such as BSE, skull stripping using GAC, and the Chan-Vese model even in the presence of noise.
In this paper a fully automatic method for segmenting MR images showing tumor, both mass-effect a... more In this paper a fully automatic method for segmenting MR images showing tumor, both mass-effect and infiltrating structures is presented. The proposed method uses UDWT and gabor wavelets. The proposed method uses T1, T2 images and produces appreciative results even in the presence of noise. A multiresolution approach using undecimated wavelet transform is employed which allows the low-low (LL), low-high (LH), high-low (HL), and high-high (HH) sub-bands to remain at full size. Detection of tumor takes place in LL. The decomposition is carried up to two levels. Gabor filters are then applied to the wavelet approximations at all levels to obtain the characteristic texture features such as entropy, second to fourth central moments and coefficient of variation. A simple peak finding algorithm is used to determine the peaks out of array of these texture features. The corresponding filter outputs are compared to obtain an image containing minimum pixel values. This is given to the kmeans clustering algorithm which then produces the final segmented output. It is observed that the algorithm captures the features from the considered levels and produces an optimal segmentation. The proposed algorithm accurately locates the tumor tissue from surrounding brain tissue.
CRC Press eBooks, Dec 9, 2013
International journal of computer applications, Jul 28, 2012
Skull stripping is an important image processing step in many neuroimaging studies. In this paper... more Skull stripping is an important image processing step in many neuroimaging studies. In this paper, a comparison of three brain extraction algorithms is done, namely Brain Surface Extractor (BSE), skull stripping algorithm using Geodesic Active Contour (GAC), and skull stripping using active contours without edges. The comparison is done with respect to accuracy of the three algorithms. The results provided by the three algorithms are compared against the processed results available in the OASIS dataset. A comparison of the three algorithms shows that BSE provides the best results with respect to the percentage of non-brain matter contained in the final segmented output. The algorithm using GAC produces a conservative result containing some amount of non-brain matter that can be removed using morphological operator. The algorithm using active contours without edges produces segmentation results containing some amount of brain matter removed from the result. This is mainly due to the sensitivity of the active contour to intensity values in the sulci present in the brain magnetic resonance image.
The application of multivariate data analysis methods such as ICA to solve the blind deconvolutio... more The application of multivariate data analysis methods such as ICA to solve the blind deconvolution problem requires the source images to be statistically independent. Since this is not always true, a subband decomposition approach is taken. Here it is assumed that the wideband source signals are dependent, but there exist some narrow subbands where they are independent. These subbands are determined by finding those subbands with minimum mutual information between corresponding nodes of the subband decomposition scheme. Subband decomposition is brought about by undecimated wavelet transform as well as Gabor wavelets. Patches are selected randomly from these subband images and given as inputs to the ICA algorithm. The ICA algorithm gives as its output the independent components which resemble short edges and capture the blurring information in the image around edges and corners. These are used as PSFs given to the blind Richardson-Lucy algorithm for deconvolution of the blurred image. The results obtained are comparable to those obtained by the blind Richardson-Lucy algorithm.
Skull stripping is an important image processing step in many neuroimaging studies. In this paper... more Skull stripping is an important image processing step in many neuroimaging studies. In this paper, a novel scheme based on a level sets representation of the geodesic active contour (GAC) is employed to detect the boundary of the skull. This approach is based on the relation between active contours and the computation of geodesics (minimal length curves). The contour is evolved from inside the MR image under the influence of geometric measures of the MR image. Before the application of GAC, the MR image is roughly done into two regions, brain and nonbrain. The centroid of the brain region is obtained which is used for drawing an ellipse situated well inside the brain region. This ellipse is the initial contour. After the model converges to a stable solution, the obtained mask is processed using morphological operators. This mask is then used to give the final segmented output.
2012 Ieee International Conference on Signal Processing Communication and Computing, Aug 1, 2012
The application of multivariate data analysis methods such as ICA to solve the blind deconvolutio... more The application of multivariate data analysis methods such as ICA to solve the blind deconvolution problem requires the source images to be statistically independent. Since this is not always true, a subband decomposition approach is taken. Here it is assumed that the wideband source signals are dependent, but there exist some narrow subbands where they are independent. These subbands are determined by finding those subbands with minimum mutual information between corresponding nodes of the subband decomposition scheme. Subband decomposition is brought about by undecimated wavelet transform as well as Gabor wavelets. Patches are selected randomly from these subband images and given as inputs to the ICA algorithm. The ICA algorithm gives as its output the independent components which resemble short edges and capture the blurring information in the image around edges and corners. These are used as PSFs given to the blind Richardson-Lucy algorithm for deconvolution of the blurred image. The results obtained are comparable to those obtained by the blind Richardson-Lucy algorithm.
Advances in Biometrics for Secure Human Authentication and Recognition, 2013
International Journal of Computer Applications, 2012
Skull stripping is an important image processing step in many neuroimaging studies. In this paper... more Skull stripping is an important image processing step in many neuroimaging studies. In this paper, a comparison of three brain extraction algorithms is done, namely Brain Surface Extractor (BSE), skull stripping algorithm using Geodesic Active Contour (GAC), and skull stripping using active contours without edges. The comparison is done with respect to accuracy of the three algorithms. The results provided by the three algorithms are compared against the processed results available in the OASIS dataset. A comparison of the three algorithms shows that BSE provides the best results with respect to the percentage of non-brain matter contained in the final segmented output. The algorithm using GAC produces a conservative result containing some amount of non-brain matter that can be removed using morphological operator. The algorithm using active contours without edges produces segmentation results containing some amount of brain matter removed from the result. This is mainly due to the sensitivity of the active contour to intensity values in the sulci present in the brain magnetic resonance image.
2012 Third International Conference on Computing, Communication and Networking Technologies (ICCCNT'12), 2012
2010 17th IEEE International Conference on Electronics, Circuits and Systems, 2010
2012 Fourth International Conference on Computational Intelligence, Communication Systems and Networks, 2012
Uploads
Papers by Gayatri Mirajkar