Projects and Research Work
Shape Reconstruction of Adipose Compartments for CT Breast Image in a Mastectomy Specimen by Ellipsoidal Approximation
Development of an efficient clinical breast imaging system has become one of the biggest concerns in the field of medical imaging. Considering the complexity of breast imaging systems, an optimized clinical imaging system needs to be developed with real breast data. The shape analysis of software breast phantoms for each adipose compartment is extended for the approximation of real breast compartment in CT image primarily collected from a mastectomy. Different slices of CT breast image are segmented at first. The number of voxels and volume of each segmented compartment are computed in ITK-SNAP. The shape analysis of real breast tissue compartments will be a true enhancement to the breast imaging simulation. The simulation parameters could be selected based on this shape and orientation information of the real breast image. Therefore, the improved realism in the generation of software breast phantoms will be achieved. IEEE SPMB 2015 [Paper, Poster] |
Big Data Representation and Clustering using the Affinity Propagation and Manifold Learning: ISOMAP
This work experiments with a tradeoff between clustering and representation. Affinity propagation is a method that solves the problem of getting stuck in local minima in Loyd's algorithm. The dataset from Stanford face data, is used to perform dimensionality reduced representation. For each representation, Affinity Propagation is performed to observe the impact of dimensionality reduced data on Affinity Propagation. |
Segmentation and Estimation of Adipose Compartments' Volume in 3-D CT Breast Images of a Mastectomy Specimen
The specimen preserved in formalin was irradiated with a high dose of X-rays and standard 3-D reconstruction techniques were undertaken. The reconstructed images were segmented in order to annotate the adipose compartments. Using ITK-SNAP, the number of voxels and the volume of each segment were computed. Subsequently, the average volume (and standard deviation) of the segmented compartments was estimated. SPIE Medical Imaging 2016 Conference [Paper, Poster] |
Validation of Partial Volume Computation using Monte Carlo Approach
We employed voxel-based approach and it was quite straightforward to compute volume of the voxels containing single tissues. But the challenge arose for finding the volumes for voxels containing multiple tissues. It could be skin with air, ligament with fat tissue, skin with dense tissue, skin with ligament and fat tissue and etc. To assess the correctness of a numerical algorithm implementation, a statistical test was developed. A Monte Carlo method was employed to estimate the accuracy of an approximation algorithm without knowing a true value to be computed. This methodology was applied in the computation of partial volumes in breast tissue simulation. TELSIKS 2015 Conference [Paper, Presentation Slides] |
Authentication in Face Recognition for Defending Against Replay Attacks
In order to defend the replay attacks by any attacker using photo of any authorized users, a facial verification technique will be incorporated with any available face recognition technique. There will be stored templates of all the authorized users in the system. A sensor in the system will capture fifty consecutive photos of the user attempting to log in system. All of the captured photos will be compared based on their histograms. If the histograms have minimum differences, then the system will not process further and notify about invasion. If the histograms of the captured images are comparatively larger, then the system will go for face recognition as face is verified. In face recognition step, one randomly chosen picture among the captured pictures will be used to find a match with any of the stored templates. If there is any matching found, then the user will be allowed to enter the system. Project Proposal PDF |
|
Improved Generation of Anthropomorphic Software Breast Phantoms
The previously developed and simulated anthropomorphic breast phantoms were regenerated using the improved parameter controlling. The octree-based algorithm was used to simulate the different ligament materials in the breast. Adipose tissue compartments, fibro-glandular tissue compartments, skin and Cooper's ligaments were present based on the defined proportion in the improved phantoms. The breast phantoms have different class of compartments on the percentage of dense tissues. |
Buzz Prediction in Online Social Media using SVM with Different Optimizations
|