OpenPros: A Large-Scale Dataset for Limited View Prostate Ultrasound Computed Tomography

TL;DR

OpenPros provides the first large-scale, anatomically faithful benchmark for limited-angle prostate ultrasound computed tomography, pairing 2D speed-of-sound phantoms with full-waveform ultrasound data derived from 3D clinical anatomies. By releasing open-source forward solvers and a diverse, clinically grounded dataset, the work enables rigorous evaluation of physics-based and deep learning inversion methods under realistic limited-view conditions. Baseline experiments show deep learning models dramatically improve inference speed and reconstruction quality over traditional approaches but still struggle with fine anatomical detail and robust generalization to unseen anatomies. This resource is poised to accelerate development of clinically viable, high-resolution prostate USCT imaging and encourages reproducible benchmarking across methods and settings.

Abstract

Prostate cancer is one of the most common and lethal cancers among men, making its early detection critically important. Although ultrasound imaging offers greater accessibility and cost-effectiveness compared to MRI, traditional transrectal ultrasound methods suffer from low sensitivity, especially in detecting anteriorly located tumors. Ultrasound computed tomography provides quantitative tissue characterization, but its clinical implementation faces significant challenges, particularly under anatomically constrained limited-angle acquisition conditions specific to prostate imaging. To address these unmet needs, we introduce OpenPros, the first large-scale benchmark dataset explicitly developed for limited-view prostate USCT. Our dataset includes over 280,000 paired samples of realistic 2D speed-of-sound (SOS) phantoms and corresponding ultrasound full-waveform data, generated from anatomically accurate 3D digital prostate models derived from real clinical MRI/CT scans and ex vivo ultrasound measurements, annotated by medical experts. Simulations are conducted under clinically realistic configurations using advanced finite-difference time-domain and Runge-Kutta acoustic wave solvers, both provided as open-source components. Through comprehensive baseline experiments, we demonstrate that state-of-the-art deep learning methods surpass traditional physics-based approaches in both inference efficiency and reconstruction accuracy. Nevertheless, current deep learning models still fall short of delivering clinically acceptable high-resolution images with sufficient accuracy. By publicly releasing OpenPros, we aim to encourage the development of advanced machine learning algorithms capable of bridging this performance gap and producing clinically usable, high-resolution, and highly accurate prostate ultrasound images. The dataset is publicly accessible at https://open-pros.github.io/.

Figures (6)

• Figure 1: OpenPros dataset creation and benchmarking pipeline. Top panel: Starting from clinical MRI and CT scans, we employ expert annotations to generate detailed 3D anatomical segmentations. We then incorporate real ultrasound speed-of-sound (SOS) measurements from ex vivo prostate samples acquired using the QTscan platform. These are integrated into comprehensive 3D abdominal SOS models. Clinically relevant 2D slices are subsequently extracted from these models to simulate limited-angle ultrasound tomography scenarios. Bottom panel: The extracted 2D SOS maps form the ground truth for ultrasound simulations governed by the acoustic wave equation. The resulting simulated ultrasound data are organized into the OpenPros dataset. We utilize these data to train and benchmark physics-based and deep-learning inversion methods, facilitating the evaluation and development of rapid, clinically relevant SOS reconstruction methods under challenging limited-angle conditions.
• Figure 2: (a) Anatomical structure and probe placement. Two probes-abdominal (on the body surface) and transrectal (in the rectum)-are used in our simulation. Image courtesy of Complete Anatomy. (b) 3D digital SOS phantom. SOS distribution in the anatomically realistic prostate model.
• Figure 3: Examples of simulated ultrasound data and phantoms: without (top) and with (bottom) bone in the phantoms. We show two example channels of reflections and transmissions with sources (yellow stars) and receivers (red dots) on the two probes. Our PDE solvers can simulate complex and realistic ultrasound wave phenomena, such like transmissions, reflections, direct waves and multi-scatterings.
• Figure 4: Benchmark results for limited‐angle prostate USCT. Each column shows a different inversion method on the same phantom: (col 1) ground‐truth SOS map; (col 2) Delay‐and‐Sum beamforming; (col 3) physics‐based USCT; (col 4) InversionNet; (col 5) ViT‐Inversion. Rows correspond to three representative prostate slices illustrating challenging (top), moderate (middle), and simple (bottom) anatomical scenarios. Zoom-in figures of the prostate region (orange squares) are shown in Figure \ref{['fig:zoom_results']}.
• Figure 5: Zoom-in comparison of prostate regions. Enlarged views (orange squares in Figure \ref{['fig:benchmark_results']}) showing detailed reconstruction quality within the prostate region across baseline methods: (col 1) Ground truth; (col 2) Delay-and-Sum beamforming; (col 3) physics-based USCT; (col 4) InversionNet; (col 5) ViT-Inversion. Each row corresponds to the same anatomical scenario as in Figure \ref{['fig:benchmark_results']}. Note that although the learned methods recover general anatomical shapes more clearly, the fine internal structures and boundaries remain poorly resolved.