Research
My research revolves around the central theme of medical imaging and image analysis, biomedical signal processing and point-of-care technology development. The work carried out so far has given me exposure to wide areas of multidimensional signal processing, physics and instrumentation of medical imaging, biomedical optics and embedded systems.
Augmemted Reality and Fluorescence Assisted Surgery
Surgeries are one of the most ancient aspects of medicine and a constant source of innovation and improvement. The AugmenTed reaLity Assisted Surgery (ATLAS) system developed by the ATLAS Consortium (Maxer Endoscopy GmbH, TUM and LMU Munich) combines an advanced navigation system, state of the art computer vision algorithms, and high-quality images of Maxer's fluorescence camera to provide a 'GPS' system to surgeons (and robots) to navigate the surgical field. The pre-operative CT/ MR/ ultrasound images of relevant and critical structures are rendered directly onto the laparoscopic video using advanced image fusion algorithms. The system uses display technologies taking the benefit of massive parallelization of Graphics Processing Units (GPUs). It enables the surgical staff to focus on the critical structure, and show structured hidden behind the surfaces. Initial user studies and pre-clinical studies demonstrate that AR-assisted surgeries can positively influence the outcome in complex surgeries and reduce stress for surgeons.
Funded by BMBF Germany | Winners of the German Medical Awards (Medical Innovations) 2020
Visual Quality Enhancement for Multi-spectral Optoacoustic Tomography (MSOT)
Multi-spectral Optoacoustic Tomography (MSOT) is capable of high resolution 3D visualization of molecular probes located deep in scattering living tissues. The primary objectives of the project are visual quality improvement of MSOT images, and enhancement of system capabilities for pre-clinical and clinical investigations.
Optimal self-calibration of tomographic reconstruction parameters:
Mismatch between the actual and predicted speed of sound values may lead to image distortions but can be mitigated by manual or automatic optimization based on metrics of image sharpness. Three new hybrid methods are suggested in this work, which are shown to outperform well-established autofocusing algorithms in mouse experiments in vivo.
Geometric Pixel Super-resolution MSOT:
We investigate a method that integrates information from multiple optoacoustic images acquired at sub-diffraction steps into one high resolution image by means of an iterative registration algorithm. The suggested approach renders significant improvements in terms of optoacoustic image resolution and quality without introducing significant alterations into the signal acquisition hardware or inversion algorithms.
History of development - Optoacoustic / Photoacoustic imaging :
Optimization of Magnetic Resonance (MR) Acquisition Parameters
Accurate quantification of brain iron would be useful in early diagnosis of Alzheimer’s disease. In this work the MR acquisition parameters for quantitative measurement of brain iron (T2) were optimized through the use of Cramér-Rao bound (CRB) analysis. The noise performance at different echo-times was analyzed at different signal-to-noise ratios and at different T2. CRB analysis shows that noise performance is independent of first echo-time and there exists a line of optimality around which the optimal second echo-times for a two echo acquisition are clustered.
This work was carried out under the mentorship of GE scientist John Schenck, one of the pioneers of high-field MRI technologies who enabled the first whole human brain scans.
Four Dimensional Volumetric Multispectral Optoacoustic (vMSOT) Imaging of Perfusion in-vivo
Intravital imaging within heterogenic solid tumors – illustrates blood perfusion profiles, hypoxic and nutrition gradients, cell viability, proliferation and drug response potentials.
vMSOT visualizes 3D anatomic, vascular and functional tumor profiles in real time.
Obtained concurrent observation from entire tumor.
Truly 4D imaging performance enabled assessment of blood oxygenation gradients and vascularization in solid breast tumors
Three types of blood perfusion profiles are revealed in breast
The method is readily adaptable to operate in a handheld clinical mode.
Machine Learning and Optoacoustic Imaging pushes the limits of Artificial Reproduction
We innovate new imaging and machine learning techniques which provides excellent molecular sensitivity, enabling the selection of competent ocytes without disrupting the follicles. Our technique improves the accurate prediction of ovarian reserve significantly and, once standardized for in vivo human imaging, could provide an effective tool for clinical infertility management.
Read more about the methods:
Non - destructive imaging of ovarian follicles [Royal Society publishing]
Grading of mammalian CoCs using Machine Learning [IEEE Xplore]
Point of Care (POC) Devices for Low Resource Regions
Cardiac Pre-screening Device using Ultra-low Power Embedded System:
The designed system is a point-of-care (POC) device that can deliver heart-care services to the rural population and bridge the rural-urban divide in healthcare delivery. The product design incorporates several innovations including the effective use of adaptive and multiresolution signal-processing techniques for acquisition, denoising, segmentation, and characterization of the heart sounds (HS) and murmurs using an ultralow-power embedded Mixed Signal Processor. The device is able to provide indicative diagnosis of cardiac conditions and classify a subject into either normal, abnormal, ischemic, or valvular abnormalities category. Preliminary results demonstrated by the prototype confirm the applicability of the device as a prescreening tool that can be used by paramedics in rural outreach programs.
Computer vision approach towards detection of Malaria Parasites:
We present an optimized normalized cut method for segmentation of RBCs infected with malarial parasites using peripheral blood smears. The algorithm is applied over various color spaces to find its optimal performance for microscopic blood smear images. We tested the efficacy of results in RGB, YCbCr, HSV and NTSC using the Rand's Index. The work is useful in telepathology applications and can automate the screening of malaria in rural areas where healthcare manpower is limited.
