Search Results (205)

Background/Objectives: Accurate diagnosis of brain tumors is one of the most important challenges in neuro-oncology since tumor classification and volumetric segmentation inform treatment planning. Two-dimensional classification and three-dimensional segmentation deep learning models can augment radiological workflows, particularly if paired with explainable AI techniques to improve model interpretability. The objective of this research was to develop a web-based brain tumor segmentation and classification diagnosis platform. Methods: A diagnosis system was developed combining 2D tumor classification and 3D volumetric segmentation. Classification employed a fine-tuned MobileNetV2 model trained on a glioma, meningioma, pituitary tumor, and normal control dataset. Segmentation employed a SegResNet model trained on BraTS multi-channel MRI with synthetic no-tumor data. A meta-classifier MLP was used for binary tumor detection from volumetric features. Explainability was offered using XRAI maps for 2D predictions and Gaussian overlays for 3D visualizations. The platform was incorporated into a web interface for clinical use. Results: MobileNetV2 2D model recorded 98.09% classification accuracy for tumor classification. 3D SegResNet obtained Dice coefficients around 68–70% for tumor segmentations. The MLP-based tumor detection module recorded 100% detection accuracy. Explainability modules could identify the area of the tumor, and saliency and overlay maps were consistent with real pathological features in both 2D and 3D. Conclusions: Deep learning diagnosis system possesses improved brain tumor classification and segmentation with interpretable outcomes by utilizing XAI techniques. Deployment as a web tool and a user-friendly interface made it suitable for clinical usage in radiology workflows. Full article

Background: Pancreatic cystic lesions (PCLs), including intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs), pose a diagnostic challenge due to their variable malignant potential. Current guidelines, such as Fukuoka and American Gastroenterological Association (AGA), have moderate predictive accuracy and may lead to overtreatment or missed malignancies. Artificial intelligence (AI), incorporating machine learning (ML) and deep learning (DL), offers the potential to improve risk stratification, diagnosis, and management of PCLs by integrating clinical, radiological, and molecular data. This is the first systematic review to evaluate the application, performance, and clinical utility of AI models in the diagnosis, classification, prognosis, and management of pancreatic cysts. Methods: A systematic review was conducted in accordance with PRISMA guidelines and registered on PROSPERO (CRD420251008593). Databases searched included PubMed, EMBASE, Scopus, and Cochrane Library up to March 2025. The inclusion criteria encompassed original studies employing AI, ML, or DL in human subjects with pancreatic cysts, evaluating diagnostic, classification, or prognostic outcomes. Data were extracted on the study design, imaging modality, model type, sample size, performance metrics (accuracy, sensitivity, specificity, and area under the curve (AUC)), and validation methods. Study quality and bias were assessed using the PROBAST and adherence to TRIPOD reporting guidelines. Results: From 847 records, 31 studies met the inclusion criteria. Most were retrospective observational (n = 27, 87%) and focused on preoperative diagnostic applications (n = 30, 97%), with only one addressing prognosis. Imaging modalities included Computed Tomography (CT) (48%), endoscopic ultrasound (EUS) (26%), and Magnetic Resonance Imaging (MRI) (9.7%). Neural networks, particularly convolutional neural networks (CNNs), were the most common AI models (n = 16), followed by logistic regression (n = 4) and support vector machines (n = 3). The median reported AUC across studies was 0.912, with 55% of models achieving AUC ≥ 0.80. The models outperformed clinicians or existing guidelines in 11 studies. IPMN stratification and subtype classification were common focuses, with CNN-based EUS models achieving accuracies of up to 99.6%. Only 10 studies (32%) performed external validation. The risk of bias was high in 93.5% of studies, and TRIPOD adherence averaged 48%. Conclusions: AI demonstrates strong potential in improving the diagnosis and risk stratification of pancreatic cysts, with several models outperforming current clinical guidelines and human readers. However, widespread clinical adoption is hindered by high risk of bias, lack of external validation, and limited interpretability of complex models. Future work should prioritise multicentre prospective studies, standardised model reporting, and development of interpretable, externally validated tools to support clinical integration. Full article

Purpose: Predicting colorectal cancer liver metastasis (CRLM) is essential for prognostic assessment. This study aims to develop and validate an interpretable multimodal machine learning framework based on multiparametric MRI for predicting CRLM, and to enhance the clinical interpretability of the model through SHapley Additive exPlanations (SHAP) analysis and deep learning visualization. Methods: This multicenter retrospective study included 463 patients with pathologically confirmed colorectal cancer from two institutions, divided into training (n = 256), internal testing (n = 111), and external validation (n = 96) sets. Radiomics features were extracted from manually segmented regions on axial T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI). Deep learning features were obtained from a pretrained ResNet101 network using the same MRI inputs. A least absolute shrinkage and selection operator (LASSO) logistic regression classifier was developed for clinical, radiomics, deep learning, and combined models. Model performance was evaluated by AUC, sensitivity, specificity, and F1-score. SHAP was used to assess feature contributions, and Grad-CAM was applied to visualize deep feature attention. Results: The combined model integrating features across the three modalities achieved the highest performance across all datasets, with AUCs of 0.889 (training), 0.838 (internal test), and 0.822 (external validation), outperforming single-modality models. Decision curve analysis (DCA) revealed enhanced clinical net benefit from the integrated model, while calibration curves confirmed its good predictive consistency. SHAP analysis revealed that radiomic features related to T2WI texture (e.g., LargeDependenceLowGrayLevelEmphasis) and clinical biomarkers (e.g., CA19-9) were among the most predictive for CRLM. Grad-CAM visualizations confirmed that the deep learning model focused on tumor regions consistent with radiological interpretation. Conclusions: This study presents a robust and interpretable multiparametric MRI-based model for noninvasively predicting liver metastasis in colorectal cancer patients. By integrating handcrafted radiomics and deep learning features, and enhancing transparency through SHAP and Grad-CAM, the model provides both high predictive performance and clinically meaningful explanations. These findings highlight its potential value as a decision-support tool for individualized risk assessment and treatment planning in the management of colorectal cancer. Full article

Background and Objectives: General-purpose multimodal large language models (LLMs) are increasingly used for medical image interpretation despite lacking clinical validation. This study evaluates the diagnostic reliability of ChatGPT-4o and Claude 2 in photographic assessment of adolescent idiopathic scoliosis (AIS) against radiological standards. This study examines two critical questions: whether families can derive reliable preliminary assessments from LLMs through analysis of clinical photographs and whether LLMs exhibit cognitive fidelity in their visuospatial reasoning capabilities for AIS assessment. Materials and Methods: A prospective diagnostic accuracy study (STARD-compliant) analyzed 97 adolescents (74 with AIS and 23 with postural asymmetry). Standardized clinical photographs (nine views/patient) were assessed by two LLMs and two orthopedic residents against reference radiological measurements. Primary outcomes included diagnostic accuracy (sensitivity/specificity), Cobb angle concordance (Lin’s CCC), inter-rater reliability (Cohen’s κ), and measurement agreement (Bland–Altman LoA). Results: The LLMs exhibited hazardous diagnostic inaccuracy: ChatGPT misclassified all non-AIS cases (specificity 0% [95% CI: 0.0–14.8]), while Claude 2 generated 78.3% false positives. Systematic measurement errors exceeded clinical tolerance: ChatGPT overestimated thoracic curves by +10.74° (LoA: −21.45° to +42.92°), exceeding tolerance by >800%. Both LLMs showed inverse biomechanical concordance in thoracolumbar curves (CCC ≤ −0.106). Inter-rater reliability fell below random chance (ChatGPT κ = −0.039). Universal proportional bias (slopes ≈ −1.0) caused severe curve underestimation (e.g., 10–15° error for 50° deformities). Human evaluators demonstrated superior bias control (0.3–2.8° vs. 2.6–10.7°) but suboptimal specificity (21.7–26.1%) and hazardous lumbar concordance (CCC: −0.123). Conclusions: General-purpose LLMs demonstrate clinically unacceptable inaccuracy in photographic AIS assessment, contraindicating clinical deployment. Catastrophic false positives, systematic measurement errors exceeding tolerance by 480–1074%, and inverse diagnostic concordance necessitate urgent regulatory safeguards under frameworks like the EU AI Act. Neither LLMs nor photographic human assessment achieve reliability thresholds for standalone screening, mandating domain-specific algorithm development and integration of 3D modalities. Full article

The growing demand for accurate and efficient Chest X-Ray (CXR) interpretation has prompted the development of AI-driven systems to alleviate radiologist workload and reduce diagnostic variability. This paper introduces the Intelligent Humanized Radiology Analysis System (IHRAS), a modular framework that automates the end-to-end process of CXR analysis and report generation. IHRAS integrates four core components: (i) deep convolutional neural networks for multi-label classification of 14 thoracic conditions; (ii) Grad-CAM for spatial visualization of pathologies; (iii) SAR-Net for anatomical segmentation; and (iv) a large language model (DeepSeek-R1) guided by the CRISPE prompt engineering framework to generate structured diagnostic reports using SNOMED CT terminology. Evaluated on the NIH ChestX-ray dataset, IHRAS demonstrates consistent diagnostic performance across diverse demographic and clinical subgroups, and produces high-fidelity, clinically relevant radiological reports with strong faithfulness, relevancy, and alignment scores. The system offers a transparent and scalable solution to support radiological workflows while highlighting the importance of interpretability and standardization in clinical Artificial Intelligence applications. Full article

Objectives: Oncocyte-rich indeterminate thyroid nodules (O-ITNs) present diagnostic and management challenges due to overlapping features between benign and malignant lesions and differing cytological classifications. This study aimed primarily to assess the ultrasound (US) characteristics and US-based risk of O-ITNs using the American College of Radiology Thyroid Imaging Reporting And Data Systems (ACR TI-RADS). A secondary objective was to compare the Bethesda System for Reporting Thyroid Cytopathology (BSRTC) and Italian Consensus for the Classification and Reporting of Thyroid Cytology (ICCRTC) cytological systems regarding classification and clinical management implications for O-ITNs. Methods: A retrospective study was conducted on 177 ITNs (TIR3A and TIR3B) evaluated between June 2023 and December 2024 at CTO-Alesini, Rome (Italy). Nodules were assessed with US, cytology, and histology. Oncocyte predominance was defined as >70% oncocytes on fine-needle aspiration (FNA). US features were analyzed according to ACR TI-RADS. Nodules were reclassified by BSRTC, and potential differences in clinical case management (CCM) were analyzed. Results: O-ITNs comprised 47.5% of the sample. Compared to non-O-ITNs, O-ITNs were larger and more frequently showed low-risk US features, including a higher prevalence of ACR TI-RADS 3 nodules. However, no progressive increase in the risk of malignancy (ROM) was observed across ACR TI-RADS classes within O-ITNs. Histological malignancy was identified in 47.1% of O-ITNs, a lower proportion compared to non-O-ITNs, though the difference was not statistically significant. Classification discordance with potential management impact was lower in O-ITNs (20.2%) than in non-O-ITNs (38.7%). Conclusions: O-ITNs typically exhibit benign-appearing US features and lower classification discordance between BSRTC and ICCRTC, yet US risk stratification fails to differentiate malignancy risk within O-ITNs. A tailored approach integrating cytology and cautious US interpretation is essential for optimal O-ITN management. Full article

Between approximately 725 and 518 Ma, a suite of specialized felsic plutons and granitic stocks were emplaced across the Arabian Shield, many of which are now recognized as highly mineralized prospects enriched in rare earth elements (REEs), rare metals, and radioactive elements bearing mineralizations. The current investigation focused on the radiological and geochemical characterization of naturally occurring radionuclides, specifically ²³⁸U, ²²⁶Ra, ²³²Th, and ⁴⁰K, within three strategically selected granitic prospects, namely, J. Tawlah albite granite (TW), J. Hamra (HM), and J. Abu Al Dod alkali feldspar syenite and granites (AD). Concerning the radioactivity levels of the investigated granitic stocks, specifically the activity concentrations of ²³⁸U, ²²⁶Ra, ²³²Th, and ⁴⁰K, the measured average values demonstrate significant variability across the TW, HM, and AD stocks. The average ²³⁸U concentrations are 195 (SD = 38.7), 88.66 (SD = 25.6), and 214.3 (SD = 140.8) Bq/kg for TW, HM, and AD granitic stocks, respectively. Corresponding ²²⁶Ra levels are recorded at 172.4 (SD = 34.6), 75.62 (SD = 25.9), and 198.4 (SD = 139.5) Bq/kg. For ²³²Th, the concentrations are markedly elevated in TW at 5453.8 (SD = 2182.9) Bq/kg, compared to 77.16 (SD = 27.02) and 160.2 (SD = 103.8) Bq/kg in HM and AD granitic stocks, respectively. Meanwhile, ⁴⁰K levels are reported at 1670 (SD = 535.9), 2846.2 (SD = 249.9), and 3225 (SD = 222.3) Bq/kg for TW, HM, and AD granitic plutons, respectively. Notably, these values exceed the global average background levels, indicating an anomalous enrichment of the studied granitic occurrences. The mean radiological hazard indices for each granitic unit generally exceed global benchmarks, except for AEDE_out in the HM and AD stocks, which remain below international limits. The geochemical disparities observed are indicative of post-magmatic alteration processes, as substantiated by the interpretation of remote sensing datasets. In light of the significant radiological burden presented by these granitic stocks, it is essential to implement a rigorous precautionary framework for any future mining. These materials must be categorically excluded from uses that entail direct human exposure, especially in residential construction or infrastructure projects. Full article

Medical Visual Question Answering (MedVQA) lies at the intersection of computer vision, natural language processing, and clinical decision-making, aiming to generate accurate responses from medical images paired with complex inquiries. Despite recent advances in vision–language models (VLMs), their use in healthcare remains limited by a lack of interpretability and a tendency to produce direct, unexplainable outputs. This opacity undermines their reliability in medical settings, where transparency and justification are critically important. To address this limitation, we propose a zero-shot chain-of-thought prompting framework that guides VLMs to perform multi-step reasoning before arriving at an answer. By encouraging the model to break down the problem, analyze both visual and contextual cues, and construct a stepwise explanation, the approach makes the reasoning process explicit and clinically meaningful. We evaluate the framework on the PMC-VQA benchmark, which includes authentic radiological images and expert-level prompts. In a comparative analysis of three leading VLMs, Gemini 2.5 Pro achieved the highest accuracy (72.48%), followed by Claude 3.5 Sonnet (69.00%) and GPT-4o Mini (67.33%). The results demonstrate that chain-of-thought prompting significantly improves both reasoning transparency and performance in MedVQA tasks. Full article

Background: Assessing pancreatic ductal adenocarcinoma (PDAC) resectability after neoadjuvant therapy (NAT) remains a diagnostic challenge. Traditional computed tomography (CT) criteria often fail to distinguish viable tumor from fibrosis, necessitating a reassessment of imaging-based standards. Methods: A comprehensive literature review was conducted using PubMed, focusing on prospective and retrospective studies over the past 25 years that evaluated the role of CT and complementary imaging modalities (MRI, PET-CT) in predicting resectability post-NAT in non-metastatic PDAC. Studies with small sample sizes or case reports were excluded. Results: Across studies, conventional CT parameters—particularly >180° vascular encasement—showed a limited correlation with histologic invasion or surgical outcomes after NAT. Persistent vessel contact on CT often reflected fibrosis, rather than active tumor. Dynamic changes, such as regression in the tumor–vessel interface and vessel lumen restoration, correlated more accurately with R0 resection. Adjunct markers like CA 19-9 response and patient performance status further improved resectability prediction. Conclusions: CT-based resectability assessment after NAT should transition from static morphologic criteria to response-based interpretation. Multidisciplinary evaluation integrating radiologic, biochemical, and clinical findings is essential to guide surgical decision-making and improve patient outcomes. Full article

Malignant diseases represent a major global health challenge and are among the leading causes of death worldwide. Accurate early diagnosis is essential for improving outcomes and combating these conditions effectively. Currently, the diagnosis of malignancies relies heavily on radiological imaging and pathological examinations, which are often invasive and not cost-effective. As such, there is a growing need for non-invasive and accessible methods to detect cancer in its early stages. Tumor markers—biomolecules whose levels increase in malignancy and can be measured in blood or other biological tissues and fluids—offer a promising tool. However, the sensitivity and specificity of currently available tumor markers are insufficient for early detection, limiting their use primarily to disease monitoring rather than diagnosis. While ongoing research continues to identify novel tumor markers, the development of more effective early detection strategies requires more than the discovery of new biomarkers. The continuous monitoring of patients and individuals with a high tumor risk and the personalization of tumor marker interpretation are also critical. In this review, we (i) summarize the most commonly used tumor markers, (ii) examine strategies for developing novel biomarkers, particularly through omics technologies, (iii) explore the potential of continuous monitoring using wearable biosensors for early tumor detection, and (iv) discuss approaches to personalizing tumor marker interpretation to support early diagnosis and improve treatment outcomes. Full article

Background: Renal tumors, encompassing benign, malignant, and normal variants, represent a significant diagnostic challenge in radiology due to their overlapping visual characteristics on computed tomography (CT) scans. Manual interpretation is time consuming and susceptible to inter-observer variability, emphasizing the need for automated, reliable classification systems to support early and accurate diagnosis. Method and Materials: We propose KidneyNeXt, a custom convolutional neural network (CNN) architecture designed for the multi-class classification of renal tumors using CT imaging. The model integrates multi-branch convolutional pathways, grouped convolutions, and hierarchical feature extraction blocks to enhance representational capacity. Transfer learning with ImageNet 1K pretraining and fine tuning was employed to improve generalization across diverse datasets. Performance was evaluated on three CT datasets: a clinically curated retrospective dataset (3199 images), the Kaggle CT KIDNEY dataset (12,446 images), and the KAUH: Jordan dataset (7770 images). All images were preprocessed to 224 × 224 resolution without data augmentation and split into training, validation, and test subsets. Results: Across all datasets, KidneyNeXt demonstrated outstanding classification performance. On the clinical dataset, the model achieved 99.76% accuracy and a macro-averaged F1 score of 99.71%. On the Kaggle CT KIDNEY dataset, it reached 99.96% accuracy and a 99.94% F1 score. Finally, evaluation on the KAUH dataset yielded 99.74% accuracy and a 99.72% F1 score. The model showed strong robustness against class imbalance and inter-class similarity, with minimal misclassification rates and stable learning dynamics throughout training. Conclusions: The KidneyNeXt architecture offers a lightweight yet highly effective solution for the classification of renal tumors from CT images. Its consistently high performance across multiple datasets highlights its potential for real-world clinical deployment as a reliable decision support tool. Future work may explore the integration of clinical metadata and multimodal imaging to further enhance diagnostic precision and interpretability. Additionally, interpretability was addressed using Grad-CAM visualizations, which provided class-specific attention maps to highlight the regions contributing to the model’s predictions. Full article

Background/Objectives: The long-term sequelae of COVID-19 pneumonia, particularly the persistence of imaging abnormalities and their relationship to clinical symptoms, remain unclear. While the acute radiologic patterns are well-documented, the transition to chronic pulmonary changes—and their implications for long COVID symptoms—require systematic investigation. Methods: Our study included 93 patients with moderate to severe COVID-19 pneumonia who were admitted to Istanbul Medical Faculty Hospital, each having one follow-up CT scan over a ten-month period. Two thoracic radiologists independently calculated semi-quantitative initial chest CT scores to evaluate lung involvement in pneumonia (0–5 per lobe, total score 0–25). Two radiologists and one pulmonologist retrospectively examined the persistence of follow-up imaging findings, interpreting them alongside the relevant clinical and laboratory data. Additionally, in a subcohort (n = 46), mid-term (5–7 months) and long-term (≥10 months) scans were compared to assess temporal trajectories. Results: Among the 93 patients with long-term follow-up imaging, non-fibrotic changes persisted in 34 scans (36.6%), while fibrotic-like changes were observed in 70 scans (75.3%). The most common persistent non-fibrotic changes were heterogeneous attenuation (29%, n = 27) and ground-glass opacities (17.2%, n = 16), and the persistent fibrotic-like changes were pleuroparenchymal bands or linear atelectasis (58%, n = 54), fine reticulation (52.6%, n = 49), and subpleural curvilinear lines (34.4%, n = 32). Both persistent non-fibrotic and fibrotic-like changes were statistically correlated with the initial CT score (p < 0.001), LDH (p < 0.001), and ferritin levels (p = 0.008 and p = 0.003, respectively). Fatigue (p = 0.025) and chest pain (p < 0.001) were reported more frequently in patients with persistent non-fibrotic changes, while chest pain (p = 0.033) was reported more frequently among those with persistent fibrotic-like changes. Among the 46 patients who underwent both mid- and long-term follow-up imaging, 47.2% of those with non-fibrotic changes (17 out of 36) and 10% of those with fibrotic-like changes (4 out of 40) exhibited regression over the long term. Conclusions: Initial imaging and laboratory findings may indicate persistent imaging findings related to long-term sequelae of COVID-19 pneumonia. Many of these persistent imaging abnormalities, particularly non-fibrotic changes seen in the mid-term, tend to lessen over the long term. A correlation exists between persistent imaging findings and clinical outcomes of long COVID-19, underscoring the need for further research. Full article

Background/Objectives: Sarcomas are a rare and heterogeneous group of malignant tumors, which makes early detection and grading particularly challenging. Diagnosis traditionally relies on expert visual interpretation of histopathological biopsies and radiological imaging, processes that can be time-consuming, subjective and susceptible to inter-observer variability. Methods: In this study, we aim to explore the potential of artificial intelligence (AI), specifically radiomics and machine learning (ML), to support sarcoma diagnosis and grading based on MRI scans. We extracted quantitative features from both raw and wavelet-transformed images, including first-order statistics and texture descriptors such as the gray-level co-occurrence matrix (GLCM), gray-level size-zone matrix (GLSZM), gray-level run-length matrix (GLRLM), and neighboring gray tone difference matrix (NGTDM). These features were used to train ML models for two tasks: binary classification of healthy vs. pathological tissue and prognostic grading of sarcomas based on the French FNCLCC system. Results: The binary classification achieved an accuracy of 76.02% using a combination of features from both raw and transformed images. FNCLCC grade classification reached an accuracy of 57.6% under the same conditions. Specifically, wavelet transforms of raw images boosted classification accuracy, hinting at the large potential that image transforms can add to these tasks. Conclusions: Our findings highlight the value of combining multiple radiomic features and demonstrate that wavelet transforms significantly enhance classification performance. By outlining the potential of AI-based approaches in sarcoma diagnostics, this work seeks to promote the development of decision support systems that could assist clinicians. Full article

Diffusion Tensor Imaging (DTI) is a crucial imaging technique for accurately assessing pathological changes in Cervical Spondylotic Myelopathy (CSM). However, the segmentation of spinal cord DTI images primarily relies on manual methods, which are labor-intensive and heavily dependent on the subjective experience of clinicians, and existing research on DTI automatic segmentation cannot fully satisfy clinical requirements. Thus, this poses significant challenges for DTI-assisted diagnostic decision-making. This study aimed to deliver AI-driven segmentation for spinal cord DTI. To achieve this goal, a comparison experiment of candidate input features was conducted, with the preliminary results confirming the effectiveness of applying a diffusion-free image (B0 image) for DTI segmentation. Furthermore, a deep-learning-based model, named SCS-Net (Spinal Cord Segmentation Network), was proposed accordingly. The model applies a classical U-shaped architecture with a lightweight feature extraction module, which can effectively alleviate the training data scarcity problem. The proposed method supports eight-region spinal cord segmentation, i.e., the lateral, dorsal, ventral, and gray matter areas on the left and right sides. To evaluate this method, 89 CSM patients from a single center were collected. The model demonstrated satisfactory accuracy for both general segmentation metrics (precision, recall, and Dice coefficient) and a DTI-specific feature index. In particular, the proposed model’s error rate for the DTI-specific feature index was evaluated as 5.32%, 10.14%, 7.37%, and 5.70% on the left side, and 4.60%, 9.60%, 8.74%, and 6.27% on the right side of the spinal cord, respectively, affirming the model’s consistent performance for radiological rationality. In conclusion, the proposed AI-driven segmentation model significantly reduces the dependence on DTI manual interpretation, providing a feasible solution that can improve potential diagnostic outcomes for patients. Full article

Artificial intelligence (AI) is reshaping radiological practice, with recent advancements in natural language processing (NLP), large language models (LLMs), and chatbot technologies opening new avenues for clinical integration. These AI-driven conversational agents have demonstrated potential in streamlining patient triage, optimizing imaging protocol selection, supporting image interpretation, automating radiology report generation, and improving communication among radiologists, referring physicians, and patients. Emerging evidence also highlights their role in decision-making, clinical data extraction, and structured reporting. While the clinical adoption of chatbots remains limited by concerns related to data privacy, model robustness, and ethical oversight, ongoing developments and regulatory efforts are paving the way for responsible implementation. This review provides a critical overview of the current and emerging applications of chatbots in radiology, evaluating their capabilities, limitations, and future directions for clinical and research integration. Full article