Research Projects

Optoacoustic Tomography

Optoacoustic imaging is a technique capable of providing molecular information about oxy-hemoglobin, deoxy-hemoglobin, lipids, glucose, protein, water and contrast agents like nanoparticles, indocyanine green and methylene blue. Imaging these molecules has direct impact on metabolic imaging, early cancer diagnostics, neuroimaging, stroke detection and cardiovascular imaging. Optoacoustic tomography is a imaging modality that uses the pulsed laser source as a input in the NIR range, which is absorbed by the tissues. The absorbed light is will result in a slight increase in the temperature, which results in increase in pressure. These pressure waves then traverse as ultrasound waves which are detected using broadband transducers. These recorded sound waves are then used to perform the reconstruction of the pressure distribution which is proportional to the absorption coefficient. The simulation of the detection of the ultrasound waves are governed by the optoacoustic wave equation. The existing methods in optoacoustic tomography are analytical in nature, hence does not provide quantitative information with less number of detectors. These analytical methods perform the image reconstruction using a time-reversal approach. Our work focuses on making in-vivo optoacoustic imaging more quantitative in nature. Further the developed approaches (both computational and experimental) will be extensively validation with histological images. Lastly, we also focus on using optoacoustic imaging for different application starting from cancer imaging to treatment monitoring, pharmocological studies and and cardiovascular disease diagnosis. Since optoacoustic tomography can perform real-time imaging with non-ionizing raditions, the technique is safe and therefore allows repeated usage during treatment monitoring.

Comparative evaluation of entropy maximization scheme with standard non-negative reconstruction using phantom data. Reconstructed OA image of star phantom using the (a) l₂-norm based reconstruction, (b) l₂-norm based reconstruction with thresholding, (c) entropy maximization based reconstruction. Absorption coefficient distribution after fluence correction using (d) l₂-norm based reconstruc- tion, (e) l₂-norm based reconstruction with thresholding, (f) entropy maximization based reconstruction. (g) shows the photograph of the phantom used, (h) line profile along the vertical red dashed line indicated in (b), (i) line profile along the horizontal blue dashed line indicated in (b). The negative values are plotted in a different colormap (a and d) for visualization and colormaps indicate quantitative values (in a.u).

Comparison of entropy maximization scheme with standard non-negative reconstruction at two different mice regions. Reconstructed optoacoustic images using the (a) l₂-norm based reconstruction, (b) l₂-norm based reconstruction with thresholding, (c) entropy based reconstruction and fluence correction (using segmented prior) of murine head region; (d) represents the magnitude of Fourier domain signal for (f); Reconstructed optoacoustic images using the (e) l₂-norm based reconstruction, (f) l₂-norm based reconstruction with thresholding, (g) entropy based reconstruction (using segmented prior) for the mouse abdominal region imaged in-vivo. (h) represents the magnitude of Fourier domain signal for (g);

Multimodal Imaging - (FMT-XCT and OPUS)

Fluorescent agents have been used for understanding various biological processes, wherein fluorescent probes binds to specific proteins (related to different diseased conditions). Fluorescence Molecular Tomography (FMT) is an imaging modality capable of localizing these fluorescent probes, thereby enabling us to understand progression of diseases and early diagnostic approaches. In FMT, light is shinned on the sample in the NIR wavelengths, which is absorbed and the fluorescent probes emits light at different wavelength (called emission wavelength). The data is collected with CCD camera at excitation and emission wavelengths to perform image reconstruction based on Born-approximation. Further XCT images of the mice region will be also acquired. Once fluorescence and XCT data is acquired, we can combine this information using two concepts namely Anisotropic Diffusion or regional Laplacian. The combination of XCT with FMT can be done using a soft prior approach, wherein the segmented XCT information is used as a prior and incorporated into FMT reconstruction using either as a regional Laplacian or an Helmholtz approach. I have worked on utilizing the FMT-XCT system for assessing asthmatic inflammation, early detection of atherosclerotic lesions and imaging integrin expression in non-small cell lung cancer in mice models. The focus of FIST lab would be to extend the utility of FMT to NIR-II wavelength regimes to enable high-resolution FMT imaging due to less scattering, and low autofluorescence signal.

I have also worked on combining information from ultrasound and optoacoustics, wherein structural information is provided by ultrasound and functional information is provided by optoacoustic. As opposed to FMT-XCT, optoacoustic-ultrasound (OPUS) does not need additional hardware, as the same ultrasound probe can be used to acquire both US and optoacoustic data, with very little electronic changes. Here, we try to use the multi-modal OPUS information to improve diagnostic accuracy. Combining this information can be done in two steps, firstly ultrasound information is registered with optoacoustic information to enable accurate localization. Secondly as explained earlier, regional Laplacian type regularization can be used to guide optoacoustic reconstruction. The reconstructed multi-modal OPUS information could possible enable accurate visualization of different diseased conditions in cancer and cardiovascular diseases.

Competition/Blocking of the ανβ3 binding site by pre-injection of Cilengitide: Transversal (A and B) and 3D volume rendering (C and D) images demonstrate a significant decrease of the Fluorescence Ratio (E) for the ανβ3 targeted NIRF probe IntegriSense680 when the mice were injected with Cilengitide 15 minutes prior to application of the NIRF probe

Cancer Theranostics

Theranostics is an emerging area of research which tries to perform diagnosis and enable accurate therapy protocols simultaneously. Theranostics has strong potential in clinics, since the surgeon will have access to molecular images during surgery and can potentially avoid recurrence problems. In the area of theranostics, I have worked on performing optoacoustic imaging along with photothermal/ photodynamic therapy regimes. Photothermal therapy relies on heating the tumor and killing it due to an increase in temperature. In contrary photodynamic therapy relies on activating a photosensitizer, which inturn creates singlet oxygen which is toxic for tumor cells. Note that traditionally phototherapies were performed using continuous wave lasers, while optoacoustic imaging was carried out with pulsed lasers. For the first time, we had demonstrated that light based therapies can be performed using pulsed lasers, and the same lasers can be used to obtain molecular imaging using optoacoustic tomography. Performing optoacoustic imaging during phototherapy can potentially enable accurate in-vivo therapeutics and avoid wound healing problems post-surgery.

In vivo photothermal therapy. a) Infrared (IR) thermal images of 4T1 tumour-bearing mice before and after laser irradiation (1.5 W cm−2, 800 nm, 6 min). Before irradiation, animals were injected with phosphate-buffered saline (PBS) intravenously or with OMVMel or OMVWT intravenously (i.v.) or intratumourally (i.t.). b) Tumour growth curves. Representative images of dissected tumours are also shown, except for OMVMel (i.t.) + L, which had nearly disappeared. Mean values and error bars are presented as mean ± SD, inter-group differences were assessed for significance using the paired t-test compared to control (***p < 0.001 vs. PBS with laser treatment; n = 4). c) Body weight of all animals was recorded during each treatment, with all animals appearing healthy throughout the study based on eating and behaviour.

Other Research Works

Magnetic Resonance guided High Intensity Focused Ultrasound (MRgHIFU)

Radiation therapy and Chemotherapy is prone to have side-effects while performing cancer treatments. Hence people are trying to explore the scenario of perfoming the treatment by focussing ultrasound (US) beams at the tumor region, resulting in a increasing temperature at the focussed region and killing the cancerous cells. An important aspect that is being studied by the community is the estimation of the temperature maps for performing better treatment and not allowing normal cells to die. Hence obtaining the high spatial and temporal resolution temperature maps in real-time acts a important step. The reconstruction of temperature maps can be done using the phase information provided by MRI technique. The various approaches to perform reconstruction is by applying constraint in time (TCR), model predictive filtering (MPF) and parallel imaging with UNFOLD. We are currently working on using iterative methods for accurate estimation of temperature maps in MRgHIFU. A good literature on this topic can be found in TCR, MPF and UNFOLD.

Comparison of temperature map reconstruction of the standard TCR with PITCR method. The reconstruction was performed using 33% and 16% of the acquired fully sampled data. Difference image is also shown for better comparison of reconstructed temperature distribution. The plots show the maximum temperature increase over time in the HIFU heating.

Diffuse Optical Tomography

Diffuse optical tomography (DOT) is one of the emerging biomedical imaging modality that could have a potential to give functional information, which will not be given by traditional imaging modalities like Computed Tomography (CT), Magnetic Resonance Imaging (MRI) or Ultrasound (US). The functional information provided by DOT is in terms of optical properties of the tissue (namely the absorption coefficient and scattering coefficient). This imaging modality is known to be non-invasive and non-ionizing. This imaging is modality is non-ionizing as the probing medium used here is the near infra-red (NIR) light, having a wavelength range of 600 nm - 1000 nm. DOT is the non-linear, ill-posed and under-determined problem, due to the physics of the problem. DOT is model using the forward problem and inverse problem. The forward problem is be modelled using a Diffusion Equation (DE) to estimate the boundary data given the internal optical properties. The modelled data is the matched with the experimentally measured data iteratively (a non-linear problem is solved in linear steps), which is also called as inverse problem. Hence, here we start with an initial guess and then move toward the solution by repeatedly solving the forward problem and inverse problem, until the change between the iterations are not more than 2%. These methods require advanced computational modelling, which tend to be computationally expensive. In the inverse problem, regularization parameter is always needed for obtaining a unique solution. We have proposed an procedure for automatically choosing a regularization parameter using a least-squares QR (LSQR) kind of approach. A model-resolution matrix based basis pursuit deconvolution approach is also studied to obtain better estimation of optical properties in DOT. A good review is available in Optical Tomography in Medical Imaging.

Comparison of the reconstructed absorption distribution using the ℓ2 , ℓ1 ,ℓp , and ℓ0 norm-based regularization scheme for the case of two small rectangular targets (top-left corner). The numerically generated data were corrupted with 1% normally distributed Gaussian noise. The 1-D cross section of reconstructed absorption distribution along the solid line of target image is shown in the bottom-right corner.

GPU computing in Medical Imaging

Graphics processing units (GPU) is the hardware components that has the capability of providing massive parallelism to perform matrix operations like matrix multiplication, solving system of equations and fourier transforms etc. These operations are very common in the field of medical imaging making it very attractive in clinic due to its low cost and small size. Hence we tried to run the advanced computational models like finite element method system and the inverse problem in DOT. The results indicated that we can achieve a speed up of around 40 times than a CPU. Also the GPU's were used to proposed a real-time temporal constraint reconstruction (RT-TCR) algorithm for MRgHIFU for providing real time temperature map distribution. These GPU's have been used in many medical image analysis techniqes like image registration, segmentation and deblurring and found to be very effective. A review on GPU usage in medical imaging is given in here,PPF AGILE.

Plots showing the computation time taken for each strategy [indicated in the legend of (a)] per single iteration versus mesh size (node number, NN) for (a) the 2-D case and (b) 3-D case.