Navchetan Awasthi

This study establishes the first large-scale, cross-system subjective benchmark for PAI by evaluating 11 full-reference (FR) and 3 no-reference (NR) IQMs against expert human perception. The objective is to identify which automated metrics align with expert rankings across heterogeneous imaging platforms and distortion regimes. A web-based subjective study was conducted with 104 participants, who performed 41,965 two-alternative forced-choice pairwise comparisons on 262 images acquired across four PAI systems. Bradley-Terry model-derived quality scores from subjective rankings were correlated with IQMs. FR metrics generally outperformed NR metrics, with strong dependence on dataset and distortion type. Information-Weighted Structural Similarity Index (IW-SSIM) achieved the highest meta-analytic agreement. Among NR methods, total variation (TV) showed comparatively stronger performance. Experts tend to prefer structural detail despite small artifacts, or exhibit a preference when picking between two types of artifacts. This work provides a benchmark for supporting better, domain-specific metrics and evaluation methods.

Breast cancer is the most frequently diagnosed malignancy among women worldwide and a leading cause of cancer-related mortality. Dynamic contrast-enhanced magnetic resonance imaging plays a central role in tumor characterization and treatment monitoring, particularly in patients receiving neoadjuvant chemotherapy. However, existing artificial intelligence models for breast magnetic resonance imaging are typically developed and evaluated using heterogeneous datasets, study populations, and assessment protocols, making direct comparison difficult and limiting understanding of model robustness across institutions and clinically relevant patient subgroups. The MAMA-MIA Challenge was designed to address these challenges by providing a standardized benchmark for the joint evaluation of primary tumor segmentation and prediction of pathologic complete response using pre-treatment magnetic resonance imaging only. The training cohort comprised 1,506 patients from multiple institutions in the United States, while evaluation was conducted on an external test set of 574 patients from three independent European centers to assess cross-continental and cross-institutional generalization. A unified scoring framework combined predictive performance with subgroup consistency across age, menopausal status, and breast density. Twenty-one international teams participated in the final evaluation phase. Results demonstrate substantial performance variability under a common external evaluation framework and reveal trade-offs between overall accuracy and subgroup fairness. The challenge provides standardized datasets, evaluation protocols, and public resources to promote the development of robust and equitable artificial intelligence systems for breast cancer imaging.

SignificanceLinear-array-based photoacoustic imaging (PAI) combines functional imaging with structural imaging from ultrasound. However, it suffers from depth-dependent optical attenuation due to surface illumination, resulting in decreased signal amplitude and image contrast with depth. Existing attenuation compensation methods often amplify noise, creating a trade-off between depth enhancement and image quality.AimWe aim to develop a deep learning method that addresses the coupled problem of optical attenuation compensation and denoising for linear-array-based PAI and test the applicability in vivo.ApproachWe propose a vision-transformer-based generative model to address this coupled problem. A diverse dataset was created using simulated data and experimental twin phantoms. Vascular twin phantoms were made by printing digital images onto polyurethane films to test the performance of the model. We trained and compared three deep learning architectures, Pix2Pix, Residual U-Net, and the proposed Transformer U-Net, using various loss functions, including adversarial, MSE, PSNR, SSIM, and a combined PSNR + SSIM. We tested the model on small animal tumor images.ResultsQuantitative evaluation shows that PSNR + SSIM loss is robust in preserving structural details and suppressing noise. Under the pre-specified SSIM + PSNR training objective, Trans U-Net achieves the highest SSIM and PSNR across noise levels on both datasets. In vivo validation using murine breast tumor models and in vivo breast imaging confirmed the model’s ability to enhance visualization of deep vascular structures without introducing noise amplification.ConclusionsThe proposed Trans U-Net effectively addresses the coupled problem of attenuation correction and denoising in handheld PAI. This method improves depth-resolved vascular imaging and is potentially useful in clinical and preclinical photoacoustic applications.

Medical imaging, a cornerstone of modern healthcare, provides a non-invasive means for diagnosing and monitoring a wide range of diseases. However, the interpretation of medical images often requires expert knowledge and can be time-consuming, leading to a growing interest in automated analysis methods.¹