Niraj Mahajan Email & Phone Number

Bengaluru, Ka, In

Bengaluru, Karnataka, India

La Jolla Shores, California, United States

• Achieved a 200+ FPS for Cine MRI reconstruction from severely undersampled Fourier data using an LSTM pipeline• Enhance the image reconstruction quality by exploiting the periodicity of Cine MRI • Deploy the method in a clinical MRI Scanner for real-time quality testing

Mumbai, Maharashtra, India

• Achieved state-of-the-art segmentation on the Breast USG dataset, with a modified U-Net architecture by using Pyramid Scene Parsing and Efficient Spatial Pyramid Blocks instead of the conventional Residual Blocks• Developed an end-to-end classifier-segmenter pipeline to classify breast cancer images using various levels of weak supervision for training the segmenter, to account for the scarcity of the segmentation labels• Extend this approach to obtain results on medical vision datasets like BRATS, Malaria Blood cell, Skin Lesions

Urbana, Illinois, United States

• Worked on building matched filtering based Convolutional Blocks under Prof. Narendra Ahuja at the Coordinated Sciences Laboratory• Condensed the computation of the convolutional blocks of Deep CNNs into cluster-based representative filters, basedon their extent of activation when convolved with different regions of input images• Generated activation values from the input by treating these representative filters as matched filters• Visualised the features extracted by these filters based on their convolutional output at various spatial locations• Analysed the Deep Convolutional block behaviour and identified the pivotal pixels in the image by tweaking thebackpropagation pipeline to compute the contribution of pixel positions to the class predictive score

Urbana, Illinois, United States

• Collaborated remotely with Prof. Narendra Ahuja, UIUC on Replacing FC layers with Nonneural Computation• Performed class-wise gaussian clustering of the convolutional block activations of a deep CNN, to replicate the computation performed on them by the FC Layers to enhance their explainability• Developed a novel algorithm based on shifting and shrinking classwise targets to achieve this formulation• Achieved at par performance with the conventional FC layers on datasets like MNIST, CIFAR10, CIFAR100 and ImageNet using the VGG16 architecture

Mumbai, Maharashtra, India

• Worked with Prof. Suyash Awate on the generation of Medical Image data using adversarial models• Proposed a novel method for generating location controlled abnormality in brain images using Controllable GANs,extendable to other medical imaging applications like pneumonia, retinopathy• Designed a rectifier model based on a Context Encoder with a Pix2Pix backbone to generate corresponding healthy images of tumors to enhance the GAN training• Optimised the Retinopathy Fundus Image generation using a two-stage pipeline by first generating the retinal blood vessel filamentary structure followed by overlaying the skeleton with the fundus data dimensionality to improve the filamentary structure generation• Incorporated a Pix2Pix Network, conditioned over the retinopathy grade to generate controlled abnormality images

Canberra, Australian Capital Territory, Australia

• Worked on the 3-D Registration of CT scan images under Prof. Mark Pickering, UNSW, Canberra• Employed ResNet in a siamese network architecture for improved prediction of the 3D registration parameters on synthetically generated images of the femur• Incorporated a half dark-channel filter algorithm for soft tissue removal in the actual knee CT images• Train the neural network to isolate the Femur and Tibia from CT images using 3D registration parameters

Mumbai, Maharashtra, India

• Worked under Prof. Sharat Chandran on an Application to Diagnose Autism• Developed an Android application to backup and recover survey data from the master application• Automated the deployment of back end on LAMP Servers, and redefined the parameters of installation• Web Hosted a back end server on a cloud based infrastructure with an android front end• Attempted the implementation of a Pytorch model for eye-tracking in Android with Tensorflow• Created a bash script for automated creation of port based Virtual Hosting on the back end

Master'S Degree, Computer Science

Bachelor Of Technology - Btech, Computer Science