MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641121138 A) filed by Dr. Theresa Cenate C F; Sujana Mallavaram; Joel Sam; and Vaishnavi S on October 09, 2026, for Slim: Sparse-Slice Implicit Memory-Efficient Reconstruction Using Coordinate Mlps.
Inventors include Dr. Theresa Cenate C F; Sujana Mallavaram; Joel Sam; and Vaishnavi S.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: Reconstructing three-dimensional medical volumes from a small number of two-dimensional cross-sectional slices is clinically important and computationally demanding. Conventional methods based on dense voxel grids and classical interpolation need large amounts of memory and may lose fine anatomical detail when only a few slices are available. In this invention, we introduce SLIM (Sparse-Slice Implicit Memory-Efficient Reconstruction), a framework that represents the target volume as a continuous implicit neural function. SLIM combines a multiresolution hash encoder with a compact coordinate-based multilayer perceptron (MLP) decoder to recover the missing volumetric information from sparse slices. The system accepts sparse axial, coronal or sagittal DICOM slices and performs preprocessing, intensity normalisation, spatial coordinate normalisation, sparse slice selection, multiresolution hash encoding, coordinate-based MLP decoding and volumetric reconstruction. The hash encoder stores spatial information as trainable feature vectors at several resolution levels, and the compact MLP converts the encoded coordinates into predicted intensity values. The hash-table size gives a configurable control over the trade-off between memory use and reconstruction quality. With a three-layer MLP of 64 neurons per layer and a multiresolution hash encoder, the complete trained model occupies approximately 4.03 MB under the default configuration. Reconstruction quality is measured with Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM) and Mean Absolute Error (MAE). Tests on DICOM brain MRI, abdominal CT and synthetic volumetric data show that SLIM gives better reconstruction fidelity than classical trilinear interpolation while using far less memory than dense voxel-based representations.
Disclaimer: Curated by HT Syndication.
