Objectives: The urgency of screening patients for the potential benefit of immunotherapy has prompted explorations of predictive biomarkers or stratification for colorectal cancer (CRC). This study aims to explore the molecular subtypes associated with immune-related genes (IRGs) in CRC and to further elucidate the underlying molecular mechanisms. Methods: IRGs were downloaded for non-negative matrix factorization (NMF) clustering. Kaplan-Meier survival analyses were used to assess survival differences. Gene set variation analysis was conducted for the exploration of potential biological mechanisms and pathways. The hub differentially expressed genes (DEGs) were selected via machine learning. Immunohistochemistry was performed to determine the expression of the hub gene and the infiltration of immune components. Results: A preliminary classification of CRC subclasses was conducted through the NMF consensus clustering. Cluster 1 was found to exhibit worse overall survival and progression-free survival than Cluster 2. The two clusters demonstrated different enriched pathways, tumor immune scores, and differential gene expression. The intersection of screening characteristic genes pertained to two genes, SPP1 and GUCA2A. Immunohistochemistry indicated that SPP1 was overexpressed in CRC tissues. Patients with high SPP1 expression demonstrated a poorer prognosis, whether at the transcriptional or protein level. Immunohistochemistry revealed that high SPP1 expression was associated with more neutrophil and less CD8 ⁺ T cell infiltration. Conclusions: NMF algorithm categorized CRC patients into two immunophenotypically distinct clusters with prognostic implications and a unique profile. SPP1 emerged as a crucial indicator for assessing tumor immune status, providing novel insights to enhance prognostic assessments and therapeutic strategies tailored to individual CRC patients.

The imaging criteria of malignant retropharyngeal lymph node (RLN) in nasopharyngeal cancer (NPC) have yet to be fully elucidated. This study aimed to establish predictive models based on ultrasound (US) and magnetic resonance (MR) characteristics for identifying malignant RLN in NPC patients after radiotherapy. 81 post-radiotherapy NPC patients with abnormal enlargement of RLN underwent endonasopharyngeal ultrasound-guided fine-needle aspirations (EPUS-FNA) to access the nature of RLN. The following features were assessed on US and MR: size, margin, vascular signal, echogenicity, enhancement signal and accompany with suspicious cervical nodes or not. A multivariate analysis was performed to screen out high-risk imaging features for recurrent RLN (RRLN), and models for the diagnosis of RRLN was constructed and tested with internal verification. We evaluated the clinical usefulness of the models through comparison of C-index and decision curve analysis. High-risk features of RRLN were heterogeneous echo (p < 0.01), vascular signal (p < 0.01) on EPUS, heterogeneous enhancement (p < 0.01) and minimum axis diameter > 10 mm (p < 0.01) on MR. The models based on the US and MR features showed good discrimination (AUC of 0.76 in the US model, 0.74 in the MR model and 0.77 in the US + MR model) and good net benefit in the validation group. Prediction models based on the US and MR features show good diagnostic performance for RRLN after radiotherapy in NPC patients. The combination of EPUS and MR may be constructed to provide prompt and reliable guidance to manage RLN.

Objectives: To benchmark the performance of DeepSeek-R1 against three other advanced AI reasoning models (GPT-4o, Qwen3, Grok-3) in automatically extracting T/N staging from esophageal cancer endoscopic ultrasound (EUS) complex medical reports, and to evaluate the impact of language (Chinese/English) and prompting strategy (with/without designed prompt) on model accuracy and robustness. Methods: We retrospectively analyzed 625 EUS reports for T-staging and 579 for N-staging, which were collected from 663 patients at the Sun Yat-sen University Cancer Center between 2018 and 2020. A 2 × 2 factorial design (Language × Prompt) was employed under a zero-shot setting. The performance of the models was evaluated using accuracy, and the odds ratio (OR) was calculated to quantify the comparative performance advantage between models across different scenarios. Results: Performance was evaluated across four scenarios: (1) Chinese with-prompt, (2) Chinese without-prompt, (3) English with-prompt, and (4) English without-prompt. In both T and N-staging tasks, DeepSeek-R1 demonstrated superior overall performance compared to the competitors. For T-staging, the average accuracy was (DeepSeek-R1 vs. GPT-4o vs. Qwen3 vs. Grok-3: 91.4% vs. 84.2% vs. 89.5% vs. 81.3%). For N-staging, the respective average accuracy was 84.2% vs. 65.0% vs. 68.4% vs. 51.9%. Notably, N-staging proved more challenging than T-staging for all models, as indicated by lower accuracy. This superiority was most pronounced in the Chinese without-prompt T-staging scenario, where DeepSeek-R1 achieved significantly higher accuracy than GPT-4o (OR = 7.84, 95% CI [4.62–13.30], p < 0.001), Qwen3 (OR = 5.00, 95% CI [2.85–8.79], p < 0.001), and Grok-3 (OR = 6.47, 95% CI [4.30–9.74], p < 0.001). Conclusions: This study validates the feasibility and effectiveness of large language models (LLMs) for automated T/N staging from EUS reports. Our findings confirm that DeepSeek-R1 possesses strong intrinsic reasoning capabilities, achieving the most robust performance across diverse conditions, with the most pronounced advantage observed in the challenging English without-prompt N-staging task. By establishing a standardized, objective benchmark, DeepSeek-R1 mitigates inter-observer variability, and its deployment provides a reliable foundation for guiding precise, individualized treatment planning for esophageal cancer patients.

Background: Whole-colon examination is crucial for patients with gynecologic oncology, which somehow presents challenges in some cases. This study is aiming to investigate the effectiveness of gastroscope applied in difficult colonoscopy among such patients. Methods: Patients with gynecologic oncology who underwent colon examination were assessed and categorized into two groups, gastroscope replacement group (GR-group) and normal colonoscopy group (NC-group). Gastroscope was applied in the challenging cases within GR-group following unsuccessful attempts for colonoscopy. The assessment encompassed various features, including body mass index (BMI), nutritional indicators, previous therapeutic history, and the number and size of gynecologic oncology. A multivariate analysis was performed to screen out high-risk features for difficult colonoscopy, and a model was developed to evaluate the appropriateness of gastroscope application in such instances. Decision curve analysis was conducted to evaluate the clinical usefulness of the model. Results: We retrospectively collected the clinical characteristics of 171 patients with gynecologic oncology who underwent colon examinations, with 19 cases and 20 cases of difficult colonoscopy in GR-group and NC-group, respectively. The success rate of passing through the narrow site was 63% for the GR-group and 60% for the NC-group (p = 1.000). High risk factors of difficult colonoscopy included a BMI ≤18 kg/m2 (p = 0.03) and post-treatment status (p < 0.01). The evaluation model, constructed from clinical features, demonstrated excellent discrimination with an area under curve (AUC) of 0.86, offering significant net benefits to the validation group. Conclusions: A low BMI and post-treatment status emerged as independent risk factors for difficult colonoscopy in patients with gynecologic oncology. In such cases, gastroscope proved to be an effective and reliable alternative for the colon examination. The current findings are promising to be put into clinical use after verification by large-scale prospective studies.