portfolio
A glimpse of the projects we've shipped recently. We build for innovation, scalability, and great UX at the same time.
A from-scratch implementation of the classic CT reconstruction algorithm: simulated Radon-transform acquisition over 181 angles, ramp filtering in the frequency domain and back-projection, validated against the Shepp-Logan phantom with a pixel-wise error map.
A side-by-side study of spatial and non-linear denoising on a noisy sagittal brain MRI — mean, Gaussian, median, bilateral and Non-Local Means — weighing noise removal against edge and texture preservation.
Reading and visualising the three file formats medical imaging runs on — a DICOM brain MR slice, a 4D NIfTI neuroimaging volume and an Analyze 7.5 X-ray — with pydicom and nibabel, the I/O layer every clinical imaging pipeline starts with.
A complete classical image-processing pipeline on a contrast-enhanced brain MRI with a visible tumour: intensity transforms, histogram equalisation, Fourier analysis, four edge detectors, threshold and morphological segmentation, and watershed separation with a distance transform.
Reconstructing a diagnostic brain MRI directly from raw multi-coil k-space data: inverse 2D Fourier transform per coil, complex-magnitude extraction and Root-Sum-of-Squares coil combination, using PyTorch-based fastMRI on the M4Raw FLAIR dataset.