Dhawan, RuchikaRuchikaDhawanAgarwal, MihirMihirAgarwalJain, ShreyansShreyansJainRuparel, HrridayHrridayRuparelShekhar, HimanshuHimanshuShekhar2025-08-292025-08-292025-01-08https://repository.iitgn.ac.in/handle/IITG2025/20409The present disclosure relates to a system (100) for super-resolution ultrasound reconstruction from sparse data. An acquisition module (102) acquires a sequence of ultrasound frames from a region containing microbubbles, wherein the dataset is sparse with significantly fewer frames than conventional acquisitions. An encoder (114) comprising convolutional blocks with convolution, batch normalization, and ReLU activation reshapes the frames to a standardized resolution and generates condensed spatial feature representations. A transformer encoder (118), equipped with positional encoding (116), processes the temporally ordered sequence to capture dependencies and spatial relationships across frames, outputting a transformed feature sequence that models bubble trajectories. A decoder (124) with transposed convolutional layers upsamples the transformed features to produce probability maps, wherein each pixel denotes the likelihood of bubble presence and highlights continuous flow paths. A mapping module (112) integrates these probability maps into a super-resolution vascular map, providing high-resolution visualization with reduced frame requirements.en-USBio-medical engineeringPatents Published[202521001823]123456789/11251