Curtise Kin Cheung Ng

Background: Advances in magnetic resonance imaging (MRI) technology, expanding clinical applications, and reduced scan times have led to increased demand for MRI services. Effective undergraduate MRI education is therefore essential to ensure that evolving workforce needs and graduate expectations can be met safely and competently. This study aimed to implement and evaluate a hands‑on MRI learning program designed for future course‑wide implementation within an Australian four‑year undergraduate medical radiation science (MRS) course. Methods: This study was informed by a previous pilot study. A revised hands‑on MRI learning program comprising two weeks of pre‑reading (Part 1) followed by a five-day hands‑on MRI learning component spanning two consecutive calendar weeks (Part 2) was developed for students who had completed second‑year instrumentation subjects in the MRS course, with Curtin Extra co‑curricular program recognition. The program was implemented at the research MRI scanning facility of the Western Australia National Imaging Facility node. All participants performed hands‑on MRI scanning during Part 2. Pre‑ and post‑program questionnaires and revised 38‑item MRI knowledge tests were used for program evaluation. The tests retained the same broad content domains and scoring approach as the pilot‑study instrument, for which acceptable internal consistency (Cronbach’s alpha = 0.79) and convergent validity (Pearson correlation = 0.40) had been demonstrated. Descriptive statistics were applied to questionnaire and test data. Changes in self‑perceived competence were analysed using Wilcoxon signed‑rank tests, while changes in knowledge test performance were assessed using paired t‑tests. An independent samples t‑test was used to compare mean test score changes with the pilot study. Results: Eight students were enrolled. Statistically significant pre- to post-program increases were observed in overall self-perceived MRI competence and MRI knowledge test performance, from 2.63 to 3.13 ( p = 0.046) and from 39.15% to 57.24% ( p = 0.002), respectively. The effect size for the latter was very large, although the wide 95% confidence interval (CI) indicated substantial uncertainty regarding its magnitude (Cohen’s d = 1.73; 95% CI: 0.58–2.83). Participants also reported highly positive perceptions across all aspects of the program (≥ 4.13/5.00), except for the supervision item, which used a different response scale and had a mean rating of 3.13, indicating supervision close to the intended “perfect setting”. The mean improvement in test scores (18.09%) did not differ significantly from that of the pilot study (27.85%; p = 0.139). Conclusions: The hands-on MRI learning program was implemented in a small voluntary cohort of undergraduate MRS students. Statistically significant pre- to post-program increases were observed in self-perceived MRI competence and MRI knowledge test performance. No statistically significant difference in mean overall MRI knowledge test score improvement was detected between this study and our pilot study. However, because this study did not include a concurrent control group, the observed increases cannot be attributed causally to the program or any individual program component. These findings provide further information for planning and evaluating future course-wide implementation but do not establish its feasibility. The suitability, scalability, resource requirements, safety, and effectiveness of compulsory course‑wide delivery require evaluation in larger studies using controlled designs. Trial registration: Trial registration: Australian New Zealand Clinical Trials Registry, ACTRN12626000824369, registered 8 July 2026. Retrospectively registered.

Background/Objectives: To date, no systematic review has focused specifically on commercially available deep learning-based image reconstruction (DLIR) software for radiation dose optimization in pediatric radiology. The purpose of this article was to systematically review original clinical studies evaluating the performance of commercial DLIR software for radiation dose optimization in pediatric radiology and to assess their methodological quality. Methods: A literature search was conducted on 28 April 2026 using seven electronic databases. The review was registered with the Open Science Framework (Registration DOI: 10.17605/OSF.IO/3JDQS). Results: Thirteen papers met the selection criteria and were included in the review. These studies evaluated five commercial computed tomography (CT) DLIR software products. Clinically achievable radiation dose reductions ranged from 11.2% to 97.9% without compromising image quality, and in some cases, even exceeding that of the reference standard, suggesting further dose reduction potential. Excluding two studies that reported substantially lower dose reductions, the clinically achievable dose reduction range was 36.0–97.9%. Furthermore, more than three-quarters of the included articles reported dose reductions of at least 50% while maintaining image quality. Improvements in study methodology were evident among papers published from 2025 onward. Conclusions: Commercial CT DLIR software can achieve substantial radiation dose reductions in pediatric CT while maintaining image quality. However, further clinical studies are needed to evaluate a broader range of commercial DLIR software, including applications in positron emission tomography and X-ray imaging. Future studies should ideally include the full pediatric age range, prospectively collected and adequately sized datasets from underrepresented geographic regions, effective dose assessment, and clinically meaningful outcome measures such as diagnostic confidence/quality/accuracy.

Background/Objectives: Existing literature reviews have not focused on the acute and late toxicities of non-magnetic resonance imaging (MRI)-guided adaptive radiation therapy (ART), compared the impacts of non-MRI-guided versus MRI-guided ART, or evaluated the effectiveness of adaptive conventional fractionated radiation therapy (CFRT) and stereotactic body radiation therapy (SBRT) in relation to toxicity in prostate cancer (PCa). The purpose of this scoping review was to systematically identify original articles and evaluate the impact of ART on toxicity in PCa in a comprehensive manner. Methods: A literature search was conducted using electronic databases on 17 June 2025, identifying 27 eligible papers. Results: The overall median toxicities of ART in PCa were 15.0% (acute grade 1 gastrointestinal (GI)), 1.0% (acute grade 2 GI), 0.0% (acute grade 3 GI), 47.1% (acute grade 1 genitourinary (GU)), 9.6% (acute grade 2 GU), 0.0% (acute grade 3 GU), 10.0% (late grade 1 GI), 2.0% (late grade 2 GI), 0.0% (late grade 3 GI), 29.7% (late grade 1 GU), 5.0% (late grade 2 GU), and 0.0% (late grade 3 GU). The choice of image guidance modality for ART does not appear to substantially influence toxicity; however, dedicated commercial ART systems may contribute to reducing toxicity to lower levels in PCa. Furthermore, the toxicity rates of adaptive CFRT and SBRT were comparable. Conclusions: Adaptive CFRT may be considered when SBRT is unsuitable for certain patients, without increasing the risk of side effects. However, further research is warranted to evaluate dedicated commercial cone-beam computed tomography (CT)- and CT-guided ART systems.

No systematic review has previously examined the application of large language models (LLMs) for generating impressions from radiology report findings. This study systematically reviews the performance of LLMs on this task and their associated evaluation methodologies. A search of seven electronic databases on 7 August 2025 identified 15 eligible papers (average quality score: 71.4%). These articles evaluated 35 LLMs, including 21 base models. The reported performance ranges were as follows: Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-1, 35.9% (Generative Pre-Trained Transformer (GPT)-4) to 69.7% (Baichuan2-13B); ROUGE-2, 13.4% (Large Language Model Meta AI (Llama)) to 52.4% (Baichuan2-13B); and ROUGE-L, 16.5% (Chat General Language Model–Medical (ChatGLM-Med)) to 63.8% (finetuned Text-to-Text Transfer Transformer (T5)). The finetuned T5 consistently demonstrated high performance, based on Bidirectional Encoder Representations from Transformers Score (BERTScore): 89.2%; BiLingual Evaluation Understudy (BLEU)-1: 65.2%; BLEU-2: 57.9%; BLEU-3: 52.5%; BLEU-4: 48.3%; Metric for Evaluation of Translation with Explicit ORdering (METEOR): 38.1%; ROUGE-1: 59.9%; ROUGE-2: 50.9%; ROUGE-L: 63.8%; and subjective metrics (clinical usability: 4.5/5.0; completeness: 4.3/5.0; conciseness: 4.3/5.0; fluency: 4.4/5.0). These results, based on 132,043 computed tomography, echocardiography, magnetic resonance imaging, and X-ray reports, indicate its strong clinical potential for assisting radiologists in impression generation through supervised finetuning rather than prompting techniques used in closed-source LLMs.

Use of artificial intelligence (AI) has become popular in radiology and radiation oncology practices. Various commercial products such as image quality enhancement and computer-aided detection and diagnosis (CAD) tools in radiology, and auto-segmentation software in radiation therapy treatment planning are available nowadays (1-5). To promote ongoing development of AI in radiology and radiation oncology, this Research Topic provides a platform to facilitate the sharing of respective AI research findings and promote evidence-based practice. This Research Topic includes four distinct studies covering both machine learning (least absolute shrinkage and selection operator [LASSO]) and deep learning (DL) techniques (convolutional neural network [CNN], generative adversarial network [GAN] and Vision Transformers [ViT]) for image synthesis, registration, segmentation, classification, CAD and radiomics. We would like to thank all authors for their contributions. The contribution by Lerch et al. (https://doi.org/10.3389/fradi.2024.1420545) demonstrates that their GAN model, DreamOn is able to synthesize images for training the ResNet-18 model for breast ultrasound image classification, resulting in more robust model performance. Their study findings are consistent with outcomes of a previous systematic review that GAN is an effective data augmentation strategy (6). For researchers who are interested in using their DreamOn model for data augmentation, the model can be obtained from https://github.com/lucle4/DreamOn.For Dumbrique et al.’s contribution (https://doi.org/10.3389/fradi.2024.1424065), a novel DL model based on fully CNNs (FCNNs) and ViTs is developed for CAD of pneumothorax. Their model outperforms those of previous studies by Abdella et al. (7), Hongyu et al. (8), Jakhar et al. (9), Malhotra et al. (10) and Mostayed et al. (11). This paves the way for implementation of their model in clinical practice.Guo et al.’s study (https://doi.org/10.3389/fonc.2024.1451414) shows that their machine learning based radiomics model is effective for luminal and non-luminal breast cancer differentiation based on T2-weighted magnetic resonance images with the highest area under the receiver operating characteristic curve value of 0.873. Although this result is encouraging, manual segmentation of tumors is employed in their study. This subsequently affects its generalizability to certain extent (12).The last contribution covered in this Research Topic is about a research platform, SenseCare developed by Wang et al. (https://doi.org/10.3389/fradi.2024.1460889). The SenseCare allows multi-institution access to a range of DL models for image registration, segmentation, classification and CAD in radiology and radiation oncology. These tools support translational research projects across different clinical domains.It is noted that the AI applications in radiology and radiation oncology are broad. In this Research Topic, several representations of these applications are demonstrated. However, other uses such as automated structured reporting, clinical decision support, image quality enhancement, reconstruction and translation as well as optimization of examination / treatment scheduling are not included (13). Radiology and radiation oncology are leading specialties in the adoption of AI applications in healthcare. As of 2021, about 70% of the 343 United States Food and Drug Administration approved AI products were developed for these two specialties (14). According to the European Network for the Assessment of Imaging in Medicine (EuroAIM) / European Society of Medical Imaging Informatics (EuSoMII) 2024 survey with 572 responses from the European Society of Radiology (ESR) members, about half of the respondents have already used AI in clinical practice (15). Hence, we would like to encourage researchers, academics, and clinicians to continue conducting AI research in radiology and radiation oncology including safe and responsible use of AI and sharing their findings (16,17). In this way, we can ensure continuous growth of our specialties and evidence-based practice for convincing more clinical centers to adopt AI and realizing its benefits in a wider context to improve patient outcomes.

As yet, no study has investigated the use of a research magnetic resonance imaging (MRI) scanner to support undergraduate medical radiation science (MRS) students in developing their MRI knowledge and practical skills (competences). The purpose of this study was to test an innovative program for a total of 10 s- and third-year students of a MRS course to enhance their MRI competences. The study involved an experimental, two-week MRI learning program which focused on practical MRI scanning of phantoms and healthy volunteers. Pre- and post-program questionnaires and tests were used to evaluate the competence development of these participants as well as the program’s educational quality. Descriptive statistics, along with Wilcoxon signed-rank and paired t-tests, were used for statistical analysis. The program improved the participants’ self-perceived and actual MRI competences significantly (from an average of 2.80 to 3.20 out of 5.00, p = 0.046; and from an average of 34.87% to 62.72%, Cohen’s d effect size: 2.53, p < 0.001, respectively). Furthermore, they rated all aspects of the program’s educational quality highly (mean: 3.90–4.80 out of 5.00) and indicated that the program was extremely valuable, very effective, and practical. Nonetheless, further evaluation should be conducted in a broader setting with a larger sample size to validate the findings of this feasibility study, given the study’s small sample size and participant selection bias.

Thank you for the opportunity to reply to the comment raised by Hårdemark

As yet, there is no systematic review focusing on benefits and issues of commercial deep learning-based auto-segmentation (DLAS) software for prostate cancer (PCa) radiation therapy (RT) planning despite that NRG Oncology has underscored such necessity. This article’s purpose is to systematically review commercial DLAS software product performances for PCa RT planning and their associated evaluation methodology. A literature search was performed with the use of electronic databases on 7 November 2024. Thirty-two articles were included as per the selection criteria. They evaluated 12 products (Carina Medical LLC INTContour (Lexington, KY, USA), Elekta AB ADMIRE (Stockholm, Sweden), Limbus AI Inc. Contour (Regina, SK, Canada), Manteia Medical Technologies Co. AccuContour (Jian Sheng, China), MIM Software Inc. Contour ProtégéAI (Cleveland, OH, USA), Mirada Medical Ltd. DLCExpert (Oxford, UK), MVision.ai Contour+ (Helsinki, Finland), Radformation Inc. AutoContour (New York, NY, USA), RaySearch Laboratories AB RayStation (Stockholm, Sweden), Siemens Healthineers AG AI-Rad Companion Organs RT, syngo.via RT Image Suite and DirectORGANS (Erlangen, Germany), Therapanacea Annotate (Paris, France), and Varian Medical Systems, Inc. Ethos (Palo Alto, CA, USA)). Their results illustrate that the DLAS products can delineate 12 organs at risk (abdominopelvic cavity, anal canal, bladder, body, cauda equina, left (L) and right (R) femurs, L and R pelvis, L and R proximal femurs, and sacrum) and four clinical target volumes (prostate, lymph nodes, prostate bed, and seminal vesicle bed) with clinically acceptable outcomes, resulting in delineation time reduction, 5.7–81.1%. Although NRG Oncology has recommended each clinical centre to perform its own DLAS product evaluation prior to clinical implementation, such evaluation seems more important for AccuContour and Ethos due to the methodological issues of the respective single studies, e.g., small dataset used, etc.

As yet, no systematic review on commercial deep learning-based auto-segmentation (DLAS) software for breast cancer radiation therapy (RT) planning has been published, although NRG Oncology has highlighted the necessity for such. The purpose of this systematic review is to investigate the performances of commercial DLAS software packages for breast cancer RT planning and methods for their performance evaluation. A literature search was conducted with the use of electronic databases. Fifteen papers met the selection criteria and were included. The included studies evaluated eight software packages (Limbus Contour, Manteia AccuLearning, Mirada DLCExpert, MVision.ai Contour+, Radformation AutoContour, RaySearch RayStation, Siemens syngo.via RT Image Suite/AI-Rad Companion Organs RT, and Therapanacea Annotate). Their findings show that the DLAS software could contour ten organs at risk (body, contralateral breast, esophagus-overlapping area, heart, ipsilateral humeral head, left and right lungs, liver, and sternum and trachea) and three clinical target volumes (CTVp_breast, CTVp_chestwall, and CTVn_L1) up to the clinically acceptable standard. This can contribute to 45.4%–93.7% contouring time reduction per patient. Although NRO Oncology has suggested that every clinical center should conduct its own DLAS software evaluation before clinical implementation, such testing appears particularly crucial for Manteia AccuLearning, Mirada DLCExpert, and MVision.ai Contour+ as a result of the methodological weaknesses of the corresponding studies such as the use of small datasets collected retrospectively from single centers for the evaluation.

Previous studies have explored use of smart glasses in telemedicine, but no study has investigated its use in teleradiography. The purpose of this study was to implement a six-month pilot program for Western Australian X-ray operators (XROs) to use smart glasses to obtain assisted reality support in their radiography practice from their supervising radiographers, and evaluate its effectiveness in terms of XROs’ competence improvement and equipment usability. Pretest–posttest design with evaluation of the XROs’ competence (including their X-ray image quality) and smart glasses usability by XROs in two remote centers and their supervising radiographers from two sites before and after the program using four questionnaire sets and X-ray image quality review was employed in this experimental study. Paired t-test was used for comparing mean values of the pre- and post-intervention pairs of 11-point scale questionnaire and image quality review items to determine any XROs’ radiography competence improvements. Content analysis was used to analyze open questions about the equipment usability. Our study’s findings based on 13 participants (11 XROs and 2 supervising radiographers) and 2053 X-ray images show that the assisted reality support helped to improve the XROs’ radiography competence (specifically X-ray image quality), with mean post-intervention competence values of 6.16–7.39 (out of 10) and statistical significances (p < 0.001–0.05), and the equipment was considered effective for this purpose but not easy to use.