S 15 - Students session 2

Author Block: N. Pangani; Tbilisi/GE
Purpose: Lack of equipment makes practicing medicine difficult for rural doctors in Georgia. Often they face critical shortages in diagnostic imaging, physicians rely on physical examination in cases of fractures or internal injuries. Medical students experienced challenging environments which helps them understand community needs and develop professional empathy. The project aimed to expose students to the daily realities of rural healthcare, advocate for improved diagnostic resources and to raise public awareness about rural health challenges.
Methods or Background: Over 3 years, students from Tbilisi Medical Academy visited four of Georgia’s high-mountain regions. They lived with rural doctors, observed health dynamics in the community. They documented their experiences with photos, videos, and reflections. After the project, the students organized an exhibition, published a journal, and offered support free online English lessons. They also set up a fund to keep community screenings going.
Results or Findings: Students reported a better understanding of doctor-patient relationships, and awareness of social factors affecting health. They recognized the situation in rural clinics and provide support for building outpatient clinics. This would help doctors and patients have access to heating every day and transportation from home to the clinic during the mountainous season. The project showcased the resilience, commitment, and flexibility of healthcare providers in resource-limited areas. It also demonstrated how student involvement can change community awareness.
Conclusion: “Six Days of a Country Doctor” shows how hands-on, student-led programs in rural areas can support professional growth, build community ties, and encourage social responsibility in future doctors. The key for students is desire. The desire to help and not abandon the patient. By blending real-world learning with advocacy and practical support, these programs benefit both education and health outcomes in rural areas.
Limitations: No limitations were identified.
Funding for this study: Funding was provided by Tbilisi Medical Academy.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: K. Abuelsamen¹, A. Sbool¹, M. Abouda², J. Al-Nawaiseh¹, Z. I. Al-Fasfous¹, R. Hattar³, J. Qawasmi¹, E. Almashtoub⁴; ¹Zarqa/JO, ²Cairo/EG, ³Irbid/JO, ⁴Balamand/LB
Purpose: This study compares the diagnostic accuracy and safety of ChatGPT-5 and Gemini-2.5 in CT and MRI imaging across multiple body systems. It aims to assess whether including clinical context improves diagnostic performance and evaluate the potential patient safety implications of missed diagnoses.
Methods or Background: A retrospective analysis was performed on 800 radiology case reports from Radiopaedia.org (400 CT and 400 MRI) covering neurological, chest, musculoskeletal, gastrointestinal, obstetric/gynaecological, and urological systems. Each case was reviewed once by ChatGPT-5 and once by Gemini-2.5 under two conditions: imaging alone and imaging plus clinical presentation. Diagnostic accuracy and errors were recorded, with missed diagnoses classified by possible consequential harm using the WHO harm severity scale. Statistical comparisons used chi-squared and t-tests, with p < 0.05 considered significant.
Results or Findings: ChatGPT-5 accuracy rose from 14.6% without clinical presentation to 34.4% with it (p < 0.001). Gemini-2.5 improved from 15.2% to 31.3% (p < 0.001). Accuracy differences between models were not significant, either without (p = 0.746) or with (p = 0.093) presentation. Missed diagnoses with potential major harm or death decreased when clinical context was added: ChatGPT from 28.5% to 22.1% and Gemini from 26.6% to 21.9% (both p < 0.001). No significant inter-model difference in harm severity was observed (p = 0.110 without; p = 0.919 with presentation).
Conclusion: Clinical context enhances diagnostic performance for both models across systems and modalities. These findings support the adjunctive role of AI chatbots in clinical workflows and highlight the necessity for prospective validation and careful integration to reduce harmful errors.
Limitations: The limitations of this study include reliance on publicly available Radiopaedia cases and variable case complexity. These factors may affect generalisability, emphasising the value of broader validation and integration into generalisability.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: E. A. Kirillova, R. Shtentsel, P. Kozlova, I. Mashchenko, G. Trufanov; St. Petersburg/RU
Purpose: In Russia, there has been no individual competition for radiology residents that comprehensively evaluates professional knowledge and essential soft skills (communication, self-presentation, creativity etc.). To address this gap, the first nationwide individual competition for diagnostic radiology residents, ROENTGENIUS, was created to identify and support the most talented young specialists, to shape diagnostic radiology leaders of tomorrow.
Methods or Background: The one-day competition comprises three sequential stages. First, candidates independently analyse complex multimodal clinical cases by reviewing DICOM files and answer open-ended questions within a limited period of time. Second, a challenge entitled «Myself in the World of Radiology» requires residents to present their personal journey in a creative format (for example, a video presentation, a stand-up performance, or an original poem). Third, participants present a clinical case prepared according to CARE Guidelines and engage in discussion with the expert radiologists of the jury. All stages are scored by the jury according to the pre-set criteria.
Results or Findings: Between 2023 and 2025, three competitions were held with a total of 48 participants from seven Russian cities. The project was supported by major national radiological societies. This format allowed participants to demonstrate and refine hard (image analysis, report preparation, competence in choosing imaging modalities, clinical reasoning) and soft skills (communication, creativity, self-presentation), while also facilitating networking with fellow residents and jury members.
Conclusion: ROENTGENIUS is the first competition for radiology residents in Russia that highlights individual performance alongside traditional team events. It promotes a new generation of radiologists who combine clinical competence with strong communication and creative skills that are essential for real-world clinical practice. This competition provides a reproducible model that can be adapted and tested in other national contexts.
Limitations: No funding was received for this study.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: M. Bajibhai; Blackburn/UK
Purpose: The study aim was to assess the impact on health-related quality of life (HRQoL) and patient-reported outcome measures (PROM) following renal tumour cryoablation.
Methods or Background: This prospective study recruited patients undergoing renal cryoablation between October 2024 and February 2025. HRQoL was assessed using the EQ-5D-5L, covering five dimensions with five severity levels and a visual analogue scale. Assessments were at baseline, day 1, day 8, and 1, 3, and 6 months; 6-month data were incomplete at abstract submission.
Results or Findings: A total of 45 patients were recruited. Median overall health scores remained stable at 80 from baseline to month 3 (range 20–100 at baseline, 20–98 at month 3). By day 8, improvements were observed across multiple domains, with mobility and self-care increasing by 30%, usual activities by 40%, and pain/discomfort by 20% compared with baseline. By month 1, only usual activities maintained a modest 20% improvement, while other domains returned to baseline. From month 3 onwards, no further significant changes were noted, suggesting that the majority of functional recovery occurs within the first 1–2 weeks post-treatment, followed by long-term stabilisation.
Conclusion: Renal cryoablation has a minimal impact on QoL with return to baseline scores for the majority of patients after one month and improvements in HRQoL outcomes for some patient
Limitations: This study has a few notable limitations. Although HRQoL outcomes were assessed up to 12 months, longer-term effects beyond one year remain unknown. HRQoL was measured using a self-reported questionnaire, which may be influenced by patient expectations or recall bias. The lack of a control group limits the ability to attribute changes solely to cryoablation, and early post-operative assessments may reflect transient effects. Finally, as a prospective, non-randomised study, causal relationships cannot be definitively established
Funding for this study: St James's Hospital
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: M. Mazuruk, M. Kacperska, P. Laba, M. Mączewski, I. Michalowska, M. Mączewski; Warsaw/PL
Purpose: Epicardial adipose tissue (EAT) is a metabolically active fat depot surrounding the myocardium. Histological analyses have shown that in end-stage heart failure with reduced ejection fraction (HFrEF) adipocyte size alterations occur within 3 mm of the myocardium. These changes correspond to higher attenuation of perimyocardial EAT on cardiac CT. Building on these findings, this study evaluated whether radiomic analysis of perimyocardial EAT can differentiate HFrEF from controls, comparing a fully automated artificial intelligence–based workflow with manual analysis.
Methods or Background: 140 patients (70 HFrEF, 70 controls) underwent cardiac CT. EAT was segmented manually (single-slice and three-slice methods) and with a convolutional neural network (CNN) focusing on a 3 mm perimyocardial layer. Radiomic features were extracted for each segmentation. Two classification models (LASSO regression and ElasticNet) were trained to differentiate HFrEF from controls. The feature space was defined either by the top 10 radiomic features (bootstrapped selection) or by a 10-dimensional latent space from an autoencoder.
Results or Findings: Automated CNN segmentation was faster (4 vs 7 minutes per case), avoided interobserver variability, and yielded the highest classification performance. With CNN segmentation, both L1 and ElasticNet models attained AUC ≈0.95, and ElasticNet achieved 91.4% accuracy (vs 80% for L1). This outperformed the manual single-slice segmentation results (AUC ~0.93, accuracy ~86%). The three-slice manual method performed worst.
Conclusion: Radiomic EAT features can robustly distinguish HFrEF from controls. A histology-inspired perimyocardial segmentation improved classification accuracy and can be fully automated using CNN segmentation with autoencoder-derived features. This reproducible approach provides high accuracy and shows promise as an imaging biomarker for HFrEF.
Limitations: The limitations of the study are single-center design, modest sample size and lack of external validation. The automatic myocardial segmentation algorithm requires refinement for severely dilated ventricles in HFrEF.
Funding for this study: This study was supported by the Agency for Medical Research (ABM) under grant no. KPOD.07.07-IW.07-0149/24.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The local ethics committee approved the study (IK.NPIA.0021.28.2026/23).

Author Block: V. Campelo, M. Silva, J. Neves Soussa, L. Santana, L. FROTA, G. Queiroz, K. Gama, F. D. Lôpo, C. Lins; Salvador/BR
Purpose: This study evaluated medical students’ perceptions and cognitive performance in radiological interpretation within a Radiological Anatomy module during the preclinical phase of medical education.
Methods or Background: This cross-sectional observational study involved third-semester medical students enrolled in the Radiological Anatomy module. A questionnaire collecting sociodemographic data, course expectations and evaluations (using a 5-point Likert scale), as well as radiological knowledge (multiple-choice questions), was administered both before and after the module. Incomplete questionnaires were excluded, and informed consent was obtained from all participants. Quantitative variables were presented as means and standard deviations. Statistical analyses included paired comparisons using the Wilcoxon signed-rank test and unpaired comparisons using the Mann–Whitney U test, depending on variable distribution. A p-value ≤ 0.05 was considered statistically significant.
Results or Findings: The sample consisted of 150 students (20±2.4 years; 57.6% female). The questionnaires demonstrated high internal consistency, with a Cronbach’s alpha of 0.93. The mean number of correct answers increased significantly from 8.9±2.8 before the module to 15.6±2.7 after (p < 0.001). Following the module, 98.5% of students rated their ability to interpret radiological images as above average or excellent. Additionally, 93.8% and 92.7% rated sessions with radiologists and the integration with other curricular components, respectively, at the same level. Furthermore, 91% of students considered the module’s contribution to their future medical practice and 94% its clinical relevance to be above average or excellent.
Conclusion: The Radiological Anatomy module during the preclinical phase of medical education facilitated significant improvements in technical skills and subjective appreciation of the teaching process, demonstrating its value for the early development of clinical competencies in medical training.
Limitations: Single-centre study limited to one academic semester.
Funding for this study: No funding was received.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Institutional review board approval (No. 7.295.056 ; CAAE: 80990524.9.0000.5544) was obtained and informed consent collected from all participants.

Author Block: C. Argüelles Llera, L. garcía del barrio, A. Ezponda Casajus; Pamplona/ES
Purpose: To evaluate AI performance in detecting pleuroparenchymal abnormalities (pneumothorax [PTX], pleural effusion [PEF], pulmonary lesion/nodule [LES], consolidation [CO], and atelectasis [AT]) on posteroanterior chest X-ray, comparing AI, an experienced thoracic radiologist, and their combination (Radiologist+AI), using the radiologist supported by AI (Radiologist_IA) as reference.
Methods or Background: In this prospective cohort, an expert radiologist reported images blinded to AI output. AI results (AI-Rad Companion Chest X-Ray, Siemens Healthineers) were recorded separately, then the radiologist reviewed AI findings. The final report (Radiologist_IA) served as reference. Sensitivity, specificity, weighted kappa (κ), and AUC were calculated. DeLong’s and McNemar’s tests were used.
Results or Findings: Agreement between AI and the radiologist was moderate, highest for PEF (κ = 0.63) and PTX (κ = 0.475). The radiologist achieved near-perfect sensitivities (~100%) and specificities (>98%). AI showed high sensitivity for PTX (100%) but lower for AT (35.3%) and CO (53.1%). For LES, AI sensitivity (0.803) exceeded the radiologist (0.727), though specificity was higher for the radiologist (0.994 vs. 0.855). Radiologist AUC was superior to AI for PEF, CO, and AT (all p < 0.001). Radiologist+AI increased LES sensitivity (0.97 vs. 0.727; p = 0.178), not significant. LES AUC was 0.8608 (radiologist) vs. 0.9121 (Radiologist+AI, p = 0.178).
Conclusion: AI shows acceptable performance in detecting pleuroparenchymal abnormalities on chest X-ray, especially PTX. Radiologist interpretation remains most reliable; AI integration enhances sensitivity, supporting its role as a clinical adjunct.
Limitations: Not applicable.
Funding for this study: None.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: C. A. B. Hofer¹, A. J. Schlemmer¹, M. Trapp², E-M. Trapp²; ¹Graz/AT, ²Klagenfurt am Wörthersee/AT
Purpose: The aim of the project “Teddydocs from Wörthersee” is to show children in a calm and fun environment how a hospital works and to take away the children’s fear of examinations with realistic rebuilds of the different imaging methods.
Methods or Background: The “Teddydocs from Wörthersee – Stuffed Animal Clinic Klagenfurt” had its premiere in July 2025. It is a joint project between KABEG (Carinthian State Hospitals Operating Company) and the Medical University of Graz.
150 preschool and elementary school children were invited to have their stuffed animals examined by medical students (“Teddydocs”) in a hospital simulation. This also included imaging stations like X-ray, MRI, CT and Ultrasound. Starting with the “Teddy Clinic” in Graz back in 2017, the imagine stations were improved and expanded. A university course was introduced specifically for this purpose. Together with volunteers and medical students new devices were planned and built.
Results or Findings: New CT and MRI scanners were crafted to show the difference between these two modalities. Ultrasound probes that were no longer needed and videos from echocardiography and renal ultrasound simulated an ultrasound examination. X-rays were simulated by illuminating images with light and projecting them onto a table. The medical students used Chat-GPT to design X-rays of stuffed animals with broken bones as well as other pathologies.
Conclusion: This newly introduced project allows children to gain insight in the common diagnostic methods and to have first encounters in a playful way. It could be an alternative or at least an addition to mock scanners. Furthermore, it can be included in health education for children of this age.
Limitations: The crafted imaging devices cannot take away the children's fear completely, but they can make an upcoming examination a little easier.
Funding for this study: No funding.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: N. A. Yosifova, G. N. Todorov, C. Bachvarov, M. Nanev, M. Tsalta-Mladenov, V. Dimitrova-Kirilova, T. Drenski; Varna/BG
Purpose: To investigate how the outcome of an endovascular
thrombectomy (EVT) after a large vessel occlusion (LVO) depends on the anatomical
characteristics of the vessels. We aimed to evaluate the efficacy of the EVT based on the present
intraprocedural complications and the postprocedural imaging findings.
Methods or Background: In this retrospective study, for a 6-month time period, all patients underwent EVT in the therapeutic window and patients with tortuosity or kinking of major extracranial vessels (planned for EVT, not accomplished) were included. Additional inclusive criterion was presence of imaging data proving LVO.
Results or Findings: Of 20 eligible patients, there were 14 (70%) achieved a favourable outcome – full patency of the previous occluded segments. In 28.5% of this subgroup we simultaneously performed carotid arterial stenting. 4 patients (20%) of the
cohort presented with significant kinking of the vessels (associated calcinosis in 5%) – in
half of them we tried EVT without successful outcome, the anatomical peculiarity of the other 2
patients could not allow us to place the aspiration devices in the occluded segment. In 10% we distinguished slightly enlarged hypodense zones – the undesirable result was due to working at the end of the therapeutic window in the first case and “multi-thrombi” state during aspiration in the second.
Conclusion: EVT in patients with common anatomy features and data for LVO is a life-saving intervention, leading to recanalization and excellent postprocedural results. The destiny of patients with anatomical idiosyncrasy depends on the complexity of the abnormality and the opportunity of reaching occluded segments with aspiration devices.
Limitations: The main limitation of this study is the number of the patients in the current cohort – 20 patients for a 6-month period due to the highly specific area of interest.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: