Solitary neurofibroma is a rare benign non-odontogenic tumor. Particularly in the oral cavity, neurogenic tumors are rare, especially if they are malignant. Neurofibromas may present either as solitary lesions or as part of the generalised syndrome of neurofibromatosis or von Recklinghausen's disease of the skin. Clinically, oral neurofibromas usually appear as pediculated or sessile nodules, with slow growth and mostly without pain. The diagnosis can be confirmed by histological examination. Neurofibromas are immunopositive for the S-100 protein, indicating its neural origin. Treatment is surgical and the prognosis is excellent. For illustration a rare case of a solitary neurofibroma in the mandible is presented.

Cystadenolymphomas (Warthin's tumors) are the second most frequent lesions of the parotid gland. Due to their benign clinical behavior, the low rates of recurrence and malignant transformation they were classified as tumor-like lesions. In addition, a polyclonal growth of the epithelial components of the tumor could be detected. Warthin's tumors occur bilateral in 7-10%, whereas a multifocal appearance is extremely rare. Even if the pathogenesis is still unclear a heterotopia of salivary tissue during embryogenesis is the most likely explanation for the origin of these tumors in the upper neck and periparotideal region. Here we present a rare case of bilateral, multifocal, extraglandular Warthin's tumors in lymph nodes of the upper neck and give a brief review of the literature. If a primary malignancy can be excluded by a careful staging procedure prior to the operation an isolated excision of the lesions of the neck is the adequate treatment.

The aim of the present study was to analyze the size of the preoperatively calculated defect volume of MRONJ patients (medication-related osteonecrosis of the jaw) and to develop a computer-aided tool based on techniques of AI (artificial intelligence) that facilitates diagnostics of MRONJ. For this purpose, on the basis of 3D (three-dimensional) data sets of 175 MRONJ patients, the outline of the necrosis was marked with the software iPlan 3.0.5 (BrainLAB AG) thus determining the respective volume. The correlation of this volume with other 31 collected parameters was analyzed retrospectively. There was a significant correlation between the defect volume on the one hand, and the MRONJ stage, the number of performed operations, and the therapy invasiveness on the other hand. Furthermore, it could be shown that in one third of the cases no defects could be recognized in the 2D imaging by panoramic radiograph, while in the 3D data sets defects already had been demarcated. On the basis of these data, a computer-aided tool based on the principles of AI was developed and validated, which might allow the automated calculation of the 3D defect extension and classification of the MRONJ cases into a ranking system. In conclusion, preparation of a 3D image can be recommended for reliably making the diagnosis “MRONJ.” Moreover, knowledge of the exact defect volume serves as a decision-making aid with regard to the choice of therapy, ensures intraoperative certitude, and allows an assessment of the prognosis.

Patients immigrated or born to immigrated parents constitute an increasing proportion of TB in Germany. This study aimed to assess the clinical presentation and outcome of immigrated TB-monoinfected (TB/HIV–) patients versus TB patients with an HIV coinfection (TB/HIV+). Ninety-three patients (36 boys/men, 57 girls/women; age range, 2–59 years) were investigated of whom 47 were TB/HIV+ and 46 TB/HIV–. TB/HIV+ patients more frequently had an African background (76.6% [36/47], P <0.001), TB/HIV– patients (63.0% [29/46]) more often had an East-European or Asian background (P <0.001). Most common symptoms included lymphadenopathy (37.6% [35/93]) and weight loss (35.5% [33/93]). In TB/HIV+ patients acute presentation was more frequent including fever (57.4% [27/47]), gastrointestinal (44.7% [21/47]), and respiratory symptoms (36.2% [17/47]). TB skin test was highly positive in all performed cases (20/20), IGRA was positive in 98.2% (55/56) cases applied. Pulmonary involvement was relatively rare (51/93 [54.8%]), especially in TB/HIV– patients (36.10% [17/46]) vs. TB/HIV+ (70.23% [36/47]; P <0.001). An infectious etiology was suspected in only 43.5% (20/46) TB/HIV– vs. 72.3% (34/47) TB/HIV+ patients (P ⩽0.005); a malignancy was suspected in (21/46 [45.7%] TB/HIV– patients vs. 12/47 [25.5%] TB/HIV+; P ⩽0.043). The diagnostic delay between first presentation to a doctor and TB diagnosis was long in all cases, especially in TB/HIV– patients (range, 0–336 weeks; median, 8 weeks vs. TB/HIV+ patients; range, 0–288 weeks; median, 0 weeks; P <0.05). TB, especially in young immigrated HIV– patients frequently presents as an atypic extrapulmonary disease, which may end up in a delayed diagnosis of up to several years.

Background/Objectives: Large language models (LLMs) show promise for clinical decision support, yet their accuracy in interpreting specialized medical guidelines remains uncertain. Retrieval-augmented generation (RAG) may enhance performance by grounding responses in authoritative knowledge bases. This study aimed to compare the accuracy, comprehensiveness, and safety of RAG-enhanced versus standard LLMs for answering clinical questions derived from the German S3 guideline for oral cavity carcinoma. Methods: We conducted a prospective, single-blind benchmark study evaluating six LLMs: one RAG-enhanced model (Custom GPT with guideline access), one consensus-based model (ConsensusGPT), and four standard models (DeepSeek-V3.2, Mistral Small 3.2, Qwen3-Next-80B, GPT-OSS-120B). Fifty clinical questions covering 17 guideline domains were presented to each model three times, yielding 900 evaluations. Three expert reviewers assessed responses using 5-point Likert scales for accuracy, comprehensiveness, and clarity, under a single-blind procedure, the effectiveness of which was tested by a pre-specified manipulation check. We then ran a paired within-model experiment in which each base model was queried with and without guideline access through a transparent, openly released retrieval pipeline, and scored every response with a condition-blind automated judge alongside deterministic retrieval metrics computed from the logs. Secondary outcomes included hallucination rates and guideline citation behavior. Inter-rater reliability was assessed using intraclass correlation coefficients (ICCs). Results: In a paired within-model design that held each base model fixed, adding transparent guideline retrieval improved accuracy—significantly in the three weaker open-weight models (Mistral, Qwen3, and GPT-OSS) and directionally in the already-strong DeepSeek and GPT-5 bases. Because a pre-specified blinding check found that experts could still identify retrieval-augmented answers with 98.5% accuracy, we anchored causal interpretation on measures that do not depend on the human raters, ranked by their independence: deterministic, log-derived retrieval metrics first, and then an automated, condition-blind LLM judge, whose agreement with the experts (Spearman ρ = 0.81, 95.7% within-one agreement) establishes shared calibration rather than independence from their bias. Deterministically from the retrieval logs, citation groundedness rose from 0% to 51–89% and retrieval recall@5 was 92%. On the judge, content-level hallucination fell from 42% to 4% and accuracy rose by a pooled +0.64 points (95% CI 0.47–0.80); the accuracy gain persisted after adjustment for response length (+0.48, 95% CI 0.22–0.73), which retrieval shortened rather than lengthened. The accuracy gain was large for weaker base models and small or non-significant for already-strong ones, whereas the hallucination and auditability gains were consistent across all models. The human ratings reproduced the judge’s accuracy effect (+0.61, 95% CI 0.49–0.74), and GPT-5 run through the transparent pipeline showed no significant difference from the proprietary Custom GPT (judge accuracy 4.48 vs. 4.58). Conclusions: Guideline retrieval yields a reproducible, largely base-independent improvement in the safety and auditability of LLM answers to clinical guideline questions, with accuracy gains concentrated in weaker base models. Because retrieval-augmented answers are recognizable to experts, rigorous evaluation should rely on rater-independent measures, and residual hallucination continues to require human oversight.