SALS-Net datasets for T cells

Dataset Overview This dataset contains high-resolution single-cell Small-Angle Light Scattering (SALS) forward diffraction snapshots used to train and validate the SALS-Net deep-learning framework. The images capture label-free single-cell structural profiles mapping primary human CD8+ T cells through distinct functional states: functional activation, progressive chronic dysfunction (exhaustion), and targeted pharmacological remodeling. The dataset is aggregated from three independent primary human donors to facilitate rigorous leave-one-donor-out (LODO) cross-validation and ensure donor-independent model generalizability. Hardware Footprint & Imaging Context All diffraction patterns were acquired using a highly cost-effective, non-destructive optofluidic setup. Single cells were hydrodynamically aligned using a viscoelastic microfluidic platform and illuminated using an economical continuous-wave laser diode. Forward scattered light was collected via an off-the-shelf configuration of optical lenses and mirrors, and captured single-shot via a scientific camera. Dataset Architecture & Directory Structure The dataset is organized into three distinct `.zip` archives, corresponding to each primary cell donor. Within each donor archive, images are isolated into designated training and testing pipelines, sub-categorized cleanly by biological functional state.File Specifications Format: Grayscale image files (.png). Resolution: Standardized and formatted for seamless deep-learning ingestion at a raw resolution of 150x150 pixels (aligned to the SALS-Net physical input layer matrix). Normalization: Raw intensity configurations. Downstream model evaluation scripts (available on GitHub) handle real-time GPU-accelerated sample-wise min-max range scaling optimization. Code & Framework Availability The complete training, batch prediction, Grad-CAM interpretability, and post-testing dashboard compilation scripts designed to process this dataset are open-source and maintained on GitHub: 👉 GitHub Repository: https://github.com/David-UniNA/SALS-Net Usage Instructions Download the SALS-Net__Dataset_Donor_X.zip archive(s) required for your execution pipeline. Unzip and place the structured Training and Testing folders inside the relative Data/ root directory of the cloned SALS-Net GitHub repository. Run SALS_Net__Training.py or SALS-Net__Testing.py to initiate network optimization.