Over the past two decades, advances in computational power and data availability combined with increased accessibility to pre-trained models have led to an exponential rise in machine learning (ML) publications. While ML may have the potential to transform healthcare, this sharp increase in ML research output without focus on methodological rigor and standard reporting guidelines has fueled a reproducibility crisis. In addition, the rapidly growing complexity of these models compromises their interpretability, which currently impedes their successful and widespread clinical adoption. In medicine, where failure of such models may have severe implications for patients’ health, the high requirements for accuracy, robustness, and interpretability confront ML researchers with a unique set of challenges. In this review, we discuss the semantics of reproducibility and interpretability, as well as related issues and challenges, and outline possible solutions to counteracting the “black box”. To foster reproducibility, standard reporting guidelines need to be further developed and data or code sharing encouraged. Editors and reviewers may equally play a critical role by establishing high methodological standards and thus preventing the dissemination of low-quality ML publications. To foster interpretable learning, the use of simpler models more suitable for medical data can inform the clinician how results are generated based on input data. Model-agnostic explanation tools, sensitivity analysis, and hidden layer representations constitute further promising approaches to increase interpretability. Balancing model performance and interpretability are important to ensure clinical applicability. We have now reached a critical moment for ML in medicine, where addressing these issues and implementing appropriate solutions will be vital for the future evolution of the field.

Objectives: Ultrasound has low sensitivity in parenchymal lesion detection compared to contrast-enhanced computed tomography (CT). In this proof-of-concept-study, we investigate whether raw ultrasound data can be used to generate CT-like images using deep learning models in order to enhance lesion detection. Methods: The k-wave ultrasound and Astra CT simulation toolkits were used to generate 2 datasets (1000 samples each) from simulated phantoms with up to 3 inclusions. The pix2pix conditional Generative adversarial network (cGAN) was trained on 800 samples per dataset, reserving the remainder for testing. Outputs were evaluated using generalized contrast-to-noise ratio (gCNR) and Structural Similarity Index (SSIM). Segmentation of B-mode alone versus B-mode with model-generated CT overlay was performed by 2 radiologists (1 board-certified and 1 resident) and both their performance and inter-observer agreement were evaluated using the Jaccard Index. Results: Model-generated CT-like images exhibited significantly improved gCNR (0.574±0.210 to 0.873±0.163) and SSIM (0.808±0.092 to 0.912±0.062) depending on phantom and inclusion type. Computed tomography-like images sometimes highlighted lesions otherwise undetectable in B-mode. The Jaccard index for 100 test samples improved significantly when segmenting Machine-Learning-augmented B-Mode compared with B-Mode images alone (0.58±0.18 to 0.69±0.16), depending on dataset and radiologist. Inter-observer agreement also improved significantly (0.74±0.18 to 0.85±0.07 for 1 dataset). Conclusions: Deep learning models can effectively translate ultrasound data into CT-like images, improving quality and inter-observer agreement, and enhancing lesion detectability, for example, by alleviating shadowing artefacts. Advances in knowledge: Generating CT-like images using raw ultrasound RF data as input to a cGAN model results in a significant improvement in lesion detectability by, for example, alleviating acoustic shadowing. With a cGAN architecture, even relatively small datasets can successfully generate CT-like images that improve lesion detectability.

Introduction: Patient-reported outcomes (PROs) are increasingly valued in oncology for capturing treatment tolerability and quality of life, and they are emerging as important data sources for precision-medicine and AI-driven clinical workflows. While the EQ-5D-5L questionnaire remains a widely used standardized instrument, dynamic electronic PROs (ePROs) collected via mobile applications generate richer, higher-frequency longitudinal data. Their alignment with established PRO measures, however, is not well-understood, limiting their integration into routine care and downstream analytic applications. In the prospective OGIPRO trial (KEK-ZH 2021-D0051), patients with HER2-positive breast cancer reported well-being and symptoms via the Medidux ePRO platform alongside weekly EQ-5D-5L assessments. In this retrospective analysis, we used linear mixed-effects modeling to examine associations between: (i) dynamic ePRO well-being and the EQ-5D-5L visual analogue scale (VAS); (ii) dynamic ePRO symptom grades and EQ-5D-5L domain sums; (iii) ePRO symptom grades and EQ-5D-5L disutility using the EQ-5D-5L value set for Germany. Materials and Methods: The analytic dataset comprised 13,699 dynamic ePRO data points (3376 well-being ratings and 10,323 symptom grades across 91 symptom types) from 53 patients, forming high-frequency longitudinal patient trajectories. Of these, 252 and 226 time-aligned observations, respectively, were used for direct comparison with EQ-5D-5L VAS and domain scores. Results: Dynamic ePRO well-being showed strong agreement with EQ-5D-5L VAS (β = 1.061, 95% CI: 1.015–1.107), with low between-patient variability. In contrast, the agreement between aggregated ePRO symptom grades and EQ-5D-5L domain sums was weaker (β = 0.404, 95% CI: 0.307–0.501) and more heterogeneous across patients. The same applied to the agreement between ePRO symptom grades and EQ-5D-5L disutility (β = 0.213; 95% CI: 0.151–0.275). Discussion: Dynamic ePRO well-being aligns closely with EQ-5D-5L VAS scores, supporting its use as a pragmatic substitute in clinical and research settings. Aggregated symptom grades, however, showed limited concordance with EQ-5D-5L domains, indicating the need for more granular analyses on larger datasets. Conclusions: Overall, dynamic ePRO systems provide validated, high-resolution longitudinal patient data and represent a scalable foundation for patient monitoring and data-driven decision support in oncology, including future AI-based precision-medicine applications.

Rib-cage morphology is a determinant of thoracic biomechanics, ventilation, and injury response, yet statistical shape models (SSMs) of the rib cage have relied on small cohorts (∼100s of individuals) imaged by clinical computed tomography, which over-represents injury and disease. We constructed a surface-based SSM of the complete 24-rib cage from 26,275 standardised whole-body magnetic resonance imaging (MRI) scans of adults aged 19–74 years from the population-based German National Cohort (NAKO). Ribs were segmented with a deep- learning pipeline (a rib-extended SPINEPS model), reconstructed as per-rib surface meshes, and brought into dense vertex-wise correspondence by Gaussian-process morphable registration in Scalismo; the aligned ensemble was summarised by generalised Procrustes analysis and principal component analysis (PCA). Fourteen per-rib geometric descriptors provided a quantitative cross- walk between the abstract PCA modes and named shape features, and associations with sex, age, body size and composition (including body-fat percentage), and smoking exposure were estimated by multivariable regression with Benjamini–Hochberg false-discovery-rate control. Shape variation was strongly concentrated: 28 modes captured 95% of the total variance, and the first three alone accounted for 69.4% (PC1, 42.6%; PC2, 16.3%; PC3, 10.5%) and admitted consistent anatomical readings – a sexually dimorphic axis (PC1), a slender-versus-stout body- habitus contrast (PC2), and a free-rib-size axis at ribs 11–12 (PC3). The sexes were nearly fully separated along PC1 (Cohen’s d = 2.52). Body mass and body-fat percentage were the dominant modifiable correlates of rib-cage shape, whereas the association with cumulative smoking exposure was comparatively small. The model is released as a population-representative geometric reference for benchmarking and morphing donor-derived finite-element human-body models and for further large-cohort shape analysis.