Xin, Zhuoyao

Objectives Uniformity artifacts caused by defective transducer elements or scanner malfunctions degrade diagnostic image quality. Traditional quality control (QC) methods, such as phantom testing and visual or quantitative image analysis, can be labor‐intensive and limited in test frequency. This study aims to develop a deep learning framework to detect uniformity artifacts and complement traditional QC. Methods Clinical median images were generated by aggregating co‐registered grayscale ultrasound images acquired by each transducer and computing median intensity values across the image stack. A pretrained ResNet‐18 model was fine‐tuned on a dataset consisting of clinical median images from linear and curvilinear transducers. The dataset was divided into training, validation, and testing subsets, ensuring no overlap between training and test transducers. To assess generalizability, the model was also evaluated on an independent test set of 396 phantom‐validated images from linear and curvilinear transducers. Results The model achieved 100% accuracy on the first dataset's test set. On the independent test set, it attained 87.4% accuracy with high sensitivity (0.84) and specificity (0.88), demonstrating robust generalization. Occlusion sensitivity maps confirmed the model's attention to uniformity artifact regions. Conclusion The deep learning framework using clinical median images demonstrated robust performance across several linear and curvilinear transducer models. It could be integrated into the clinical QC workflow by automating artifact detection in an effective and timely manner. In our practice, it can flag median images classified as artifact‐present, reducing human review time by approximately 80% while preserving detection accuracy.

Salivary glands are common organs at risk in both head and neck external beam radiotherapy (EBRT) and radiopharmaceutical therapy (RPT), but incidences of xerostomia in RPT are inconsistent with the EBRT Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC) limits. In EBRT, salivary glands are usually assumed to be parallel organs, with QUANTEC guidelines based on Dmean, but this is known to be a gross over-simplification of the full complexity of the underlying functional organization. The goal of this work is to combine machine learning of EBRT dose–outcome data with stylized small-scale RPT dosimetry to discover more reliable normal tissue complication probability (NTCP) models of xerostomia across both modalities. A retrospective cohort of 211 EBRT patients was analyzed using a custom-designed in-house machine learning workflow. From this, a hierarchy of three models of increasing complexity was trained, evaluated for performance and generalization, and coupled with stylized small-scale salivary gland dosimetry to assess the influence of model complexity on the predicted NTCP for plausible patterns of RPT dose nonuniformity. The three models in the hierarchy (A, B, C), in increasing order of complexity, associate xerostomia with the following: the mean dose to the whole contralateral parotid (model A), the mean dose to a ductally localized region (model B) and a serial interaction dose term between two ductal sub-compartments (model C). While the difference between the three models for EBRT p-values and AUCs is rather marginal, for physiologically driven ductal dose distributions in RPT, the predicted reduction in TD50 can be as large as a factor of 10. These results provide hints towards a plausible reconciliation of the observed inconsistency of xerostomia in RPT with EBRT dose limits.