Backprojection_database

Backprojection Database for Ultrasound Computed Tomography Anomaly Detection This repository contains the Backprojection Database, a collection of 10,618 images designed for training and validating deep learning models applied to Ultrasound Computed Tomography (USCT). The dataset includes both synthetic and experimental reconstructions obtained through unfiltered backprojection, with and without structural anomalies.The images represent handwritten characters (a, b, e) under normal and artificially altered conditions, emulating internal discontinuities in soft tissues. Additionally, numerical simulations using the k-Wave toolbox and experimental reconstructions acquired with a custom robotic ultrasound system and tissue-mimicking materials are included. This combination provides a controlled yet realistic testbed for multi-class classification, anomaly localization, and validation of hybrid deep learning architectures.The dataset is linked to the research work Hybrid Deep Transfer Learning Framework for Anomaly Detection in Ultrasound Computed Tomography, currently under preparation for scientific publication. It is published on Zenodo with a DOI to ensure traceability, reproducibility, and independent citation.Format and Structure:• Images organized by class and condition (normal/anomalous).• Subfolders for synthetic data, numerical simulations, and experimental reconstructions.• Annotation files for classification and detection tasks.Applications:• Training and evaluation of CNNs for multi-class classification.• Validation of anomaly detectors in low-SNR scenarios.• Comparative studies between simulated and experimental USCT data.