OBJECTIVE:To compare diagnostic performance and fine-needle aspiration (FNA) decision making for thyroid nodules classified as "intermediate suspicion" across three major ultrasound-based risk stratification systems (2017 ACR Thyroid Imaging Reporting and Data System [ACR-TIRADS], 2015 American Thyroid Association (ATA) guidelines, and 2020 Chinese Thyroid Imaging Reporting and Data System [C-TIRADS]), with and without artificial intelligence (AI) assistance. METHODS:This retrospective study analyzed 1,911 ultrasound images of thyroid nodules from 1,040 patients (mean age, 46.1 ± 13.1 years) collected between 2021 and 2022. The intermediate suspicion category was defined as ACR-TIRADS level TR4 ("moderately suspicious"), ATA level 4, and "moderate suspicion" in C-TIRADS (level 4B). Seven radiologists independently categorized nodules according to each guideline, and an AI model independently evaluated all nodules. AI-assisted diagnostic and FNA strategies were implemented and compared across the three guidelines. Diagnostic performance and 95% confidence intervals were assessed using patient-level clustered logistic generalized estimating equations. RESULTS:In the intermediate suspicion category, diagnostic accuracy was lower for both radiologists and the AI model than in the overall cohort. Radiologists' overall accuracy increased from 66.4% to 71.4% without AI to 74.1% to 79.4% with AI across all three guidelines (P < .001). With AI, C-TIRADS achieved the highest accuracy (79.4%), specificity (66.7%), and positive predictive value (77.4%) (all P < .05), with no significant differences between ACR-TIRADS and ATA guidelines. Without AI, FNA rates were lowest with ACR-TIRADS (37.7%) and highest with C-TIRADS (52.2%). AI assistance reduced FNA rates by 20.3% to 34.7%, increased malignant detection rate by 15.0% to 23.4%, and decreased missed malignancy rate by 23.7% to 36.8% (all P < .001). With AI assistance, ATA level 4 had the lowest FNA rate (14.7%), whereas moderate suspicion in C-TIRADS (level 4B) had the lowest missed malignancy rate (29.8%) and the numerically highest malignant detection rate (70.0%). DISCUSSION:AI-assisted interpretation improves diagnostic accuracy and FNA decision making for intermediate suspicion nodules. C-TIRADS combined with AI shows superior performance among the three systems.

e16188 Background: The prognosis of advanced biliary tract cancer (BTC) remains unsatisfactory despite first-line treatment with PD-1/PD-L1 inhibitors in combination with gemcitabine and cisplatin. Lenvatinib (LEN), an anti-angiogenic multi-kinase inhibitor, has recently demonstrated promising antitumor activity in various tumors. This study aimed to determine the efficacy and safety of envafolimab (PD-L1 inhibitor) and LEN in combination with gemcitabine and cisplatin as first-line treatment in advanced BTCs. Methods: This prospective, open-label, single-arm, Simon’s two-stage, phase II study in patients with advanced BTC is ongoing at two centers. All patients received subcutaneous envafolimab (400mg, Q3W) and LEN (8 mg/day, orally once daily) combined with gemcitabine (1000mg/m2, IV, days 1, 8, Q3W) plus cisplatin (25mg/m2, IV, days 1, 8, Q3W) up for 6-8 cycles, followed by maintenance envafolimab and LEN until disease progression (PD) or unacceptable toxicity. The primary endpoint was objective response rate (ORR) per Response Evaluation Criteria in Solid Tumors 1.1 (RECIST 1.1). A Simon’s two-stage phase II optimal design was used with following statistical assumptions: If 4 or more patients had partial response (PR) in 10 patients in stage 1, the cohort would expand to a total of 39 patients, and the outcomes would be positive if 15 or more patients achieved PR. The secondary endpoints included overall survival (OS), progression-free survival (PFS), disease control rate (DCR), and safety. Results: Between October 2022 and October 2023, 20 patients were enrolled, and included for efficacy and safety analysis. All patients received at least two doses of study medication. Of the first 10 patients enrolled, confirmed responses were noted in 4 patients, and the trial continued to accrual. The ORR and DCR were 45% and 80%, respectively. Nine patients (9/20, 45%) exhibited PR, seven patients (7/20, 35%) showed stable disease (SD), and four patients (4/20, 20%) experienced PD. The most common grade 3/4 treatment-related adverse events (TRAEs) were GGT increased (n = 3, 15%) and platelet decreased (n = 2, 10%). No unacceptable toxicity or treatment-related deaths occurred. Conclusions: Envafolimab and LEN in combination with gemcitabine and cisplatin seems to be an effective and tolerable treatment option in advanced BTC. Survival data is immature and will be updated in the future. Clinical trial information: NCT05410197 .

BACKGROUND & AIMS:Local ablation triggers anti-tumor response and is regarded as an encouraging treatment combined with immunotherapy for hepatocellular carcinoma (HCC). Irreversible electroporation (IRE) is a novel ablative technique eliminating tumor cells by electroporation, however, the characteristics of IRE-induced immune microenvironment and underlying mechanism remain unclear. METHODS:We developed an orthotopic immunocompetent HCC mouse model and performed incomplete IRE-ablation. The post-IRE immune microenvironment was characterized by RNA sequencing, single-cell RNA sequencing, flow cytometry, and multiplex immunofluorescence. Cytokine-chemokine array and organoid-immune cell co-culture were used to investigate the functions and underlying mechanism. The combination therapy of IRE and anti-PD-1 was examined for HCC treatment in mice. RESULTS:IRE initially inhibited HCC growth, but tumors rapidly regrew by day 14 after ablation, with increased Ly6G+ Polymorphonuclear (PMN) Myeloid-Derived Suppressor Cells (MDSCs) infiltration and CD8+ T cell exhaustion. In patient-derived HCC organoids, IRE-induced CD8+ T cell cytotoxicity was suppressed by MDSCs. Ly6G antibody-mediated PMN-MDSC depletion repressed HCC regrowth after IRE-ablation in vivo (1.29±0.12 vs 0.49±0.09 g, p<0.001). Mechanistically, HCC cells degraded CXCL10 mRNA via YTH N6-methyladenosine RNA binding protein F2 (YTHDF2), which activated MDSCs that led to CD8+ T cells exhaustion. Combined anti-PD-1 and IRE reduced tumor burden by 84.3% compared with IRE alone (0.22±0.08 vs 1.43±0.13 g, p<0.01), prolonged survival (day-60 survival: 93.33% vs 40%, p<0.001), and induced long-term anti-tumor immunity in mice. Four unresectable recurrent HCC patients treated with IRE plus anti-PD-1 therapy showed improved recurrence-free time than patients receiving IRE alone. CONCLUSIONS:PMN-MDSCs mediated immunosuppression that promoted HCC regrowth after incomplete IRE ablation. Combining anti-PD-1 therapy is a promising approach to eliminate post-ablation residual HCC, supporting evaluation in prospecitve clinical trials. IMPACT AND IMPLICATIONS:Post-ablation recurrence remains a major challenge in HCC. Here, we show that although IRE initially activates CD8+ T cells but subsequently promotes PMN-MDSCs activation and CD8+ T-cell exhaustion through YTHDF2-mediated CXCL10 mRNA degradation, resulting in a suppressive tumor immune microenvironment (TIME). These findings identify a previously unrecognized mechanism of post-IRE tumor regrowth and provide a rationale for early combination with anti-PD-1 therapy. Although further prospective clinical validation is required, this work suggests that monitoring post-ablation immune changes and applying rational immunotherapy combinations may help optimize treatment strategies for patients with HCC undergoing local ablation.

Hepatocellular carcinoma (HCC) ultrasound screening encounters challenges related to accuracy and the workload of radiologists. This retrospective, multicenter study assessed four artificial intelligence (AI) enhanced strategies using 21,934 liver ultrasound images from 11,960 patients to improve HCC ultrasound screening accuracy and reduce radiologist workload. UniMatch was used for lesion detection and LivNet for classification, trained on 17,913 images. Among the strategies tested, Strategy 4, which combined AI for initial detection and radiologist evaluation of negative cases in both detection and classification phases, outperformed others. It not only matched the high sensitivity of original algorithm (0.956 vs. 0.991) but also improved specificity (0.787 vs. 0.698), reduced radiologist workload by 54.5%, and decreased both recall and false positive rates. This approach demonstrates a successful model of human-AI collaboration, not only enhancing clinical outcomes but also mitigating unnecessary patient anxiety and system burden by minimizing recalls and false positives.

The response of the kidney after induction treatment is one of the determinants of prognosis in lupus nephritis, but effective predictive tools are lacking. Here, we sought to apply deep learning approaches on kidney biopsies for treatment response prediction in lupus nephritis. Patients who received cyclophosphamide or mycophenolate mofetil as induction treatment were included, and the primary outcome was 12-month treatment response, complete response defined as 24-h urinary protein under 0.5 g with normal estimated glomerular filtration rate or within 10% of normal range. The model development cohort included 245 patients (880 digital slides), and the external test cohort had 71 patients (258 digital slides). Deep learning models were trained independently on hematoxylin and eosin-, periodic acid-Schiff-, periodic Schiff-methenamine silver- and Masson's trichrome-stained slides at multiple magnifications and integrated to predict the primary outcome of complete response to therapy at 12 months. Single-stain models showed area under the curves of 0.813, 0.841, 0.823, and 0.862, respectively. Further, integration of the four models into a multi-stain model achieved area under the curves of 0.901 and 0.840 on internal validation and external testing, respectively, which outperformed conventional clinicopathologic parameters including estimated glomerular filtration rate, chronicity index and reduction in proteinuria at three months. Decisive features uncovered by visualization for model prediction included tertiary lymphoid structures, glomerulosclerosis, interstitial fibrosis and tubular atrophy. Our study demonstrated the feasibility of utilizing deep learning on kidney pathology to predict treatment response for lupus patients. Further validation is required before the model could be implemented for risk stratification and to aid in making therapeutic decisions in clinical practice.

570 Background: Extrahepatic cholangiocarcinoma (ECC) and gallbladder cancer (GBC), as the majority of biliary tract cancer (BTC), has a markedly high risk of recurrence after surgery. However, adjuvant treatments specifically for resectable ECC and GBC patients are scare and adjuvant chemotherapy alone delivers limited efficacy. Immunotherapy and radiotherapy are potential effective treatments and both of them may synergize with chemotherapy. Methods: ACCORD was a multicenter, phase 2, randomized controlled trial to assess the efficacy and safety of chemoradiation with immunotherapy as an adjuvant treatment in resectable ECC/GBC, compared to observation. The primary endpoint was overall survival (OS) and the secondary endpoints included recurrence-free survival (RFS) and safety. Patients in the chemoradiation-anti-PD-1 group received Camrelizumab intravenously every 3 weeks after surgery. Camrelizumab therapy was not terminated until disease progression, unacceptable toxic effects occurred or informed consent withdrawal. After 2 courses of Camrelizumab treatment, chemoradiation was carried out simultaneously. Patients in the observation group received no anticancer treatment unless relapse was detected. Results: From March 2020 to June 2022, a total of 93 ECC/GBC patients after curative resection were randomized 1:1 into chemoradiation-anti-PD-1 group ( n =46, Camrelizumab + concurrent Capecitabine and radiotherapy) and observation group ( n =47). The 1-year, 2-year and 3-year OS rate were 95.7%, 71.4%, and 58.2% in the chemoradiation-anti-PD-1 group, and 80.9%, 52.9%, 30.5% in the observation group (Hazard ratio (HR) 0.43, 95% confidence interval 0.24-0.79; P =0.004). The 1-year, 2-year and 3-year RFS rate were 78.3%, 54.0%, and 40.3% in the chemoradiation-anti-PD-1 group, and 55.3%, 27.0%, 17.2% in the observation group, with a hazard ratio of 0.46 (95% CI, 0.28 to 0.76; P < 0.001). In the chemoradiation-anti-PD-1 group, the main adverse effect ≥ grade 3 were anemia (7 [15.2%]), dermatitis radiation (5 [10.9%]), and nausea (5 [10.9%]); and 100% of patients completed the whole treatment. Conclusions: Chemoradiation combined with immunotherapy as an adjuvant therapy demonstrated superior survival outcomes over observation in resectable ECC/GBC patients with a well-tolerable safety profile, supporting the potential of this combination treatment as effective adjuvant therapy for these high-risk patients. Clinical trial information: NCT04333927 .

BACKGROUND:Lymph node metastasis (LNM) poses a considerable threat to survival in lung adenocarcinoma. Currently, minor resection is the recommended surgical approach for small-diameter lung cancer. The accurate preoperative identification of LNM in patients with small-diameter lung cancer is important for improving patient survival and outcomes. METHODS:A total of 1740 patients with clinical early-stage lung adenocarcinoma who underwent surgical resection were enrolled in this study. The Lasso model was used to screen clinical and imaging features, and multivariate logistic regression analysis was used to analyze the relevant diagnostic factors to establish a diagnostic model for predicting LNM. Receiver operating characteristic (ROC) curve analysis, decision curve analysis (DCA) and calibration curve analysis were used to verify the clinical efficacy of the model, which was further validated with an internal validation set. RESULTS:The proportion of solid components (PSC), sphericity, nodule margin, entropy, and edge blur were identified as diagnostic factors that were strongly correlated with LNM in lung adenocarcinoma patients. The area under the ROC curve (AUC) in the internal training set was 0.91. Decision curve analysis revealed that the model could achieve greater benefits for patients. The calibration curve was used to further verify the applicability of the prediction model. CONCLUSIONS:Patients with early-stage lung adenocarcinoma with LNM can be identified by typical imaging features. The diagnostic model can help to optimize surgical planning among thoracic surgeons.

In hepatocellular carcinoma (HCC), lenvatinib is a key first-line treatment that significantly improves survival in some patients with advanced stage. However, lenvatinib resistance presents a major clinical challenge. This study aims to identify key molecular factors driving lenvatinib resistance in HCC and propose intervention strategies to overcome this resistance, thereby enhancing therapeutic efficacy. A genome-wide CRISPR-Cas9 activation screen identified METTL8 as a crucial gene associated with lenvatinib resistance. Validation through in vitro and in vivo assays confirmed METTL8’s role in mediating lenvatinib resistance. Higher METTL8 expression was observed in lenvatinib-resistant HCC cells compared to parental cells. Immunohistochemical staining of tissue sections from HCC patients revealed a negative correlation between high METTL8 expression and lenvatinib sensitivity. To inhibit the function of METTL8 that mediate lenvatinib resistance, we conducted a screening using a natural compound library, virtual drug screening identified Rabdosiin as a potential METTL8 inhibitor, subsequent experiments demonstrated that Rabdosiin could effectively overcome METTL8-mediated lenvatinib resistance. In conclusion, this research highlights METTL8 as a novel target for mitigating lenvatinib resistance, proposing that targeting METTL8 could restore lenvatinib sensitivity in HCC, and underscores its value as a biomarker for lenvatinib application in clinical settings.

BACKGROUND:Glucocorticoids are recommended for the induction and remission phase of ulcerative colitis (UC). Early identification of glucocorticoid therapy response contributes to more precise treatment management. We aim to use deep learning model to predict glucocorticoid response prognosis in active UC. METHODS:From January 2006 to December 2023, 485 intestinal histological whole slide images (WSIs) of 212 UC patients from two medical centers in China was collected. We developed and validated a deep learning model (UCG-SwinT) based on WSI and clinical data to predict the treatment response of glucocorticoid induction therapy. Response was defined as steroid effectiveness and steroid dependence. We used area under the curves (AUCs) to evaluate the performance of the model and compared it to clinical factors. Grad-CAM was used to visualize the histological features the model focused when predicting treatment response. RESULTS:The AUCs of predicting response in training, validation, and external testing set were 0.750, 0.727, and 0.723, respectively. The UCG-SwinT model performs better while combining histopathological images with clinical data than simply inputting histopathological images, with AUCs of 0.826, 0.731, and 0.725 in predicting treatment response in the training, validation, and external testing cohorts and outperformed all clinical factors. Grad-CAM showed that increased inflammatory cells and intestinal mucosal microvascular dilation are related to glucocorticoid response in UC patients. CONCLUSIONS:UCG-SwinT has the potential to predict glucocorticoid response in active UC patients and has guiding significance for individualized clinical treatment.

This file describes supplementary methods and materials.

Primary liver cancer, a common malignant tumor of the digestive tract, ranks fifth in global cancer incidence and shows high morbidity and mortality. Liver cancer patients who are diagnosed early have the option of surgical resection, which offers the possibility of a radical cure. However, due to the insidious disease onset, most patients are diagnosed in the intermediate or advanced stages, and surgery is no longer a viable option. Therefore, systemic treatment options play an essential role in the management of advanced liver cancer. These treatments aim to suppress disease progression, prolong survival, and improve quality of life. This article reviews the latest research in the field of systemic therapy of liver cancer, including molecular targeted therapy, immunotherapy, and their combination strategies. At first, the application and efficacy of first-line molecularly targeted drugs are discussed. Next, the revolutionary advances in immune checkpoint blockers are presented. Subsequently, the clinical effects of the combination of molecularly targeted therapy and immunotherapy are analyzed. Finally, this article summarizes the current challenges faced by the systemic treatment of liver cancer and introduces the prospect of future treatment trends.

Extrahepatic cholangiocarcinoma (EHC) and gallbladder cancer (GBC), which make up most biliary tract cancers, have distinct molecular and clinical features compared with intrahepatic cholangiocarcinoma. However, effective adjuvant treatments specifically for patients with resectable EHC and GBC are scarce. To evaluate the safety and efficacy of immunotherapy combining with chemoradiotherapy. In the ACCORD randomized clinical trial, from April 2020 to June 2022, patients with EHC and GBC after curative resection were assessed for eligibility. Patients were randomized 1:1 into to the combination camrelizumab plus concurrent capecitabine and radiotherapy group or the observation group. Data were analyzed from June to August 2024. The intervention group received camrelizumab every 3 weeks after surgery. After 2 courses of camrelizumab treatment, patients received capecitabine with concurrent radiotherapy. Patients in the observation group received no anticancer treatment unless relapse was detected. The primary end point was overall survival (OS) and the secondary end points included recurrence-free survival (RFS) and safety. Of 93 included patients, 48 (52%) were female, and the median (range) age was 62 (31-70) years. Patients' baseline characteristics were comparable in the 2 groups. With a median (IQR) follow up of 36 (32-39) months, patients in the combination treatment group significantly better OS and RFS. The 1-year, 2-year and 3-year OS rates were 95.7% (95% CI, 83.7-98.9), 71.4% (95% CI, 56.4-82.5), and 58.2% (95% CI, 40.4-72.4), respectively, in the combination treatment group and 80.9% (95% CI, 66.4-89.5), 52.9% (95% CI, 37.7-65.9), and 30.5% (95% CI, 16.5-45.7), respectively, in the observation group (hazard ratio, 0.43; 95% CI, 0.24-0.79; P = .004). The 1-year, 2-year and 3-year RFS rates were 78.3% (95% CI, 63.4-87.7), 54.0% (95% CI, 38.6-67.1), and 40.3% (95% CI, 25.3-54.8), respectively, in the combination treatment group and 55.3% (95% CI, 40.1-68.1), 27.0% (95% CI, 15.2-40.3), and 17.2% (95% CI, 7.7-29.8), respectively, in the observation group (hazard ratio, 0.46; 95% CI, 0.28-0.76; P < .001). In the combination treatment group, only 6 patients (13%) experienced treatment delay for camrelizumab, and all patients completed the chemoradiation treatment, with no treatment-related deaths. In this randomized clinical trial, camrelizumab plus concurrent capecitabine and radiotherapy as an adjuvant therapy demonstrated superior survival outcomes over observation in patients with resectable EHC/GBC with a well-tolerable safety profile. The observed camrelizumab plus concurrent capecitabine and radiotherapy efficacy warrants further study with active treatment (chemotherapy or chemoradiation therapy) as the control group. ClinicalTrials.gov Identifier: NCT04333927.

BackgroundTargeted therapy for intrahepatic cholangiocarcinoma (ICC) shows superior survival outcomes but patients with certain targetable alterations are no more than 20%. Genetic alteration screening for all ICC patients is of high cost and not routinely performed. This study intends to develop a histopathology-based artificial intelligence (AI)-assisted system for predicting genetic alteration of ICC.MethodsWe constructed a Genetic Alteration Prediction (GAP) system based on multi-instance learning and self-supervised learning to predict genetic alterations using whole-slide images (WSIs) of H&E-stained slides. A total of 2069 WSIs from 232 ICC patients underwent surgery of the FAH-SYSU dataset were used for model construction and adjustment by five-fold cross-validation. Another 150 patients from three medical centres were used as independent external validations. We also compared the cost-effectiveness of GAP-assisted precise treatment and all-sequencing strategy to non-sequencing strategy.ResultsThe GAP was able to predict actionable genetic alterations of ICC, including FGFR2 and IDH. The area under the receiver operating characteristic curves (AUC) for FGFR2 and IDH were 0.754 and 0.713 in the internal dataset, and 0.724 and 0.656 in the external dataset, respectively. Furthermore, compared to giving chemotherapy without sequencing for every patient, GAP-assisted precise treatment could increase 1 progression-free quality-adjusted life month with a cost of $13871.72, the co-responding figure for all-sequencing strategy is $44538.93. Decision curve analysis showed that AI-assisted strategy provides better clinical benefits.ConclusionsWe constructed an AI-assisted genetic alteration screening system which is predictable to ICC actionable targets and has potential to assist precise targeted treatment of advanced ICC.

BACKGROUND:Postoperative pulmonary complications (PPCs) contribute to high mortality rates and impose significant financial burdens. In this study, a machine learning-based prediction model was developed to identify patients at high risk of developing PPCs following laparoscopic hepatectomy. METHODS:Data were collected from 1022 adult patients who underwent laparoscopic hepatectomy at two centres between January 2015 and February 2021. The dataset was divided into a development set and a temporal external validation set based on the year of surgery. A total of 42 factors were extracted for pre-modelling, including the implementation status of Enhanced Recovery after Surgery (ERAS). Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) method. Model performance was assessed using the area under the receiver operating characteristic curve (AUC). The model with the best performance was externally validated using temporal data. RESULTS:The incidence of PPCs was 8.7%. Lambda.1se was selected as the optimal lambda for LASSO feature selection. For implementation of ERAS, serum gamma-glutamyl transferase levels, malignant tumour presence, total bilirubin levels, and age-adjusted Charleston Comorbidities Index were the selected factors. Seven models were developed. Among them, logistic regression demonstrated the best performance, with an AUC of 0.745 in the internal validation set and 0.680 in the temporal external validation set. CONCLUSIONS:Based on the most recent definition, a machine learning model was employed to predict the risk of PPCs following laparoscopic hepatectomy. Logistic regression was identified as the best-performing model. ERAS implementation was associated with a reduction in the number of PPCs.

BACKGROUND:The liver is the most common site of metastasis from gastrointestinal stromal tumors (GISTs). The authors aimed to evaluate imatinib (IM) combined with hepatic resection (HR) or other local treatments such as radiofrequency ablation (RFA) and transarterial chemoembolization (TACE), compared to IM monotherapy in long-term survival benefits in patients suffering from GIST liver metastases.METHODS:Our research encompassed 238 patients diagnosed with liver metastases of GISTs from January 2002 to April 2022 at the First Affiliated Hospital of Sun Yat-Sen University. The oncological outcomes of concern included overall survival (OS), progression-free survival (PFS), and liver-specific PFS.RESULTS:Of all 238 patients, 126 were treated with IM alone (IM group), 81 with IM combined with HR (IM+HR group), and 31 with IM combined with RFA/TACE (IM+RFA/TACE group). The median follow-up time was 44.83 months. The median OS in the IM group was 132.60 months and was not reached in either the IM+HR group or the IM+RFA/TACE group. The 10-year OS rate in the IM+HR group was significantly superior to the IM group and the IM+RFA/TACE group (91.9% vs. 61.1% vs. 55.2%, respectively, P =0.015), and the liver-specific PFS ( P =0.642) and PFS ( P =0.369) in the three groups showed a beneficial trend in the combined treatment group. Multivariate analyses showed that age less than or equal to 60 years (HR 0.280, P< 0.001) and IM+HR (HR 0.361, P =0.047) were independently associated with better OS. Achieving no evidence of disease through surgical intervention was independently correlated with enhanced OS (HR 0.099, P =0.034), liver-specific PFS (HR 0.388, P =0.014), and PFS (HR 0.402, P =0.004).CONCLUSIONS:In patients with GIST liver metastases, IM combined with HR might improve OS in selected patients compared with IM alone and IM combined with RFA/TACE. Achieving no evidence of disease status with surgical treatment of patients results in significant prolonging of OS, liver-specific PFS, and PFS.

Purpose:Predicting the pathological response after neoadjuvant conversion therapy for initially unresectable hepatocellular carcinoma (HCC) is essential for surgical decision-making and survival outcomes but remains a challenge. We aimed to develop a radiomics model to predict pathological responses. Methods:We included 203 patients with HCC who underwent hepatectomy after neoadjuvant conversion therapy between 2015 and 2023 and separated them into a training set (100 patients from Center A) and a validation set (103 patients from Center B). Pathological complete response (pCR)-related radiomic features were extracted from the largest tumor layer in the arterial and portal vein phases of the CT. A synthetic minority oversampling technique (SMOTE) was used to balance the minority groups in the training set. The SMOTE radiomics model was constructed using a logistic regression model in the SMOTE training set and its performance was verified in the validation set. Results:The AUC of the preoperative modified response evaluation criteria in solid tumors (mRECIST) assessment for pCR was 0.656 and 0.589 in the training and validation sets, respectively. The SMOTE radiomics model was established based on ten radiomic features and showed good pCR-predictive performance in the SMOTE training set (AUC, 0.889; accuracy, 87.7%) and the validation set (AUC: 0.843, accuracy: 86.4%). The RFS of the radiomics-predicted-pCR group was significantly better than that of the predicted-non-pCR group in the training cohort (P = 0.001, 2-year RFS: 69.5% and 30.1% respectively) and the validation cohort (P = 0.012, 2-year RFS: 65.9% and 38.0% respectively). Conclusion:The SMOTE radiomics model has great potential for predicting pathological response and evaluating RFS in patients with unresectable HCC after neoadjuvant conversion therapy.