Talks and presentations

Conference Proceedings: Improving Mitosis Detection Via UNet-based Adversarial Domain Homogenizer

February 18, 2023

Conference Proceedings, 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023), Lisbon, Portugal

Conference proceedings on our paper: Improving Mitosis Detection Via UNet-based Adversarial Domain Homogenizer

Abstract:
The effective counting of mitotic figures in cancer pathology specimen is a critical task for deciding tumor grade and prognosis. Automated mitosis detection through deep learning-based image analysis often fails on unseen patient data due to domain shifts in the form of changes in stain appearance, pixel noise, tissue quality, and magnification. This paper proposes a domain homogenizer for mitosis detection that attempts to alleviate domain differences in histology images via adversarial reconstruction of input images. The proposed homogenizer is based on a U-Net architecture and can effectively reduce domain differences commonly seen with histology imaging data. We demonstrate our domain homogenizer’s effectiveness by showing a reduction in domain differences between the preprocessed images. Using this homogenizer with a RetinaNet object detector, we were able to outperform the baselines of the 2021 MIDOG challenge in terms of average precision of the detected mitotic figures.

Conference proceedings talk on Simulating Ultrasound Images from CT Scans

February 18, 2023

Conference proceedings talk, 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023), Lisbon, Portugal

Abstract:
Anatomical information in ultrasound (US) imaging has not been exploited fully because its wave interference pattern (WIP) has been viewed as speckle noise. We tested the idea that more information can be retrieved by disentangling the WIP rather than discarding it as noise. We numerically solved the forward model of generating US images from computed tomography (CT) images by solving wave-equations using the Stride library. By doing so, we have paved the way for using deep neural networks to be trained on the data generated by the forward model to simulate the solution of the inverse problem, which is generating the CT-style and CT-quality images from a real US image. We demonstrate qualitative features of the generated images that are rich in anatomical details and realism.

Conference Proceedings: WSSAMNet: Weakly Supervised Semantic Attentive Medical Image Registration Network

October 18, 2022

Conference Proceedings, Singapore, Singapore

Conference proceedings on our paper: “WSSAMNet: Weakly Supervised Semantic Attentive Medical Image Registration Network”

Abstract:
We present WSSAMNet, a weakly supervised method for medical image registration. Ours is a two step method, with the first step being the computation of segmentation masks of the fixed and moving volumes. These masks are then used to attend to the input volume, which are then provided as inputs to a registration network in the second step. The registration network computes the deformation field to perform the alignment between the fixed and the moving volumes. We study the effectiveness of our technique on the BraTSReg challenge data against ANTs and VoxelMorph, where we demonstrate that our method performs competitively.

Conference Proceedings: Perceptual CGAN for MRI Super-resolution

July 15, 2022

Conference Proceedings, The 44th International Engineering In Medicine and Biology Conference (EMBC), 2022, Glasgow, UK

Conference proceedings on our paper: Perceptual CGAN for MRI Super-resolution

Abstract:
Capturing high-resolution magnetic resonance (MR) images is a time consuming process, which makes it unsuitable for medical emergencies and pediatric patients. Low-resolution MR imaging, by contrast, is faster than its high-resolution counterpart, but it compromises on fine details necessary for a more precise diagnosis. Super-resolution (SR), when applied to low-resolution MR images, can help increase their utility by synthetically generating high-resolution images with little additional time. In this paper, we present an SR technique for MR images that is based on generative adversarial networks (GANs), which have proven to be quite useful in producing sharp-looking details in SR. We introduce a conditional GAN with perceptual loss, which is conditioned upon the input low-resolution image, which improves the performance for isotropic and anisotropic MRI super-resolution.

Interaction of Qualcomm Innovation Fellowship winners with Secretary of MEITY and Principal Scientific Advisor

July 29, 2021

QIF talks, Online, India

Qualcomm Innovation Fellowship winners announced - Interact with Secretary of MEITY and Principal Scientific Advisor

Abstract:
Prostate cancer is the second largest cause of cancer deaths in men worldwide. The prostate is a dense organ that is surrounded by many other organs. Consequently, it is difficult to locate suspicious regions within the prostate for a diagnostic biopsy whose needle cannot be guided easily using real-time and portable imaging techniques, such as ultrasound (US) imaging. The standard diagnostic procedure is random needle biopsy, in which there is always a chance to miss a small tumor mass of high grade. On the other hand, the tumor contours are better visible in magnetic resonance imaging (MRI). Therefore, a fusion of the two imaging modalities has the potential to combine their advantages. However, the existing solutions require manual matching of fiducial points on the two modalities, and they do not simulate real-time organ displacement and deformation during biopsy or surgery. Furthermore, the potential of the US imaging has not been exploited very well till now. For example, the wave interference pattern in the US, which is considered as “speckle noise,” contains a lot of information about the structure and pathobiology of the organ. But, this information is lost and left unused when the US images are despeckled. Instead, the information in the wave interference pattern may be usable, if disentangled.