A radiomics-based interpretable model to predict the pathological grade of pancreatic neuroendocrine tumors.

OBJECTIVES: To develop a computed tomography (CT) radiomics-based interpretable machine learning (ML) model to predict the pathological grade of pancreatic neuroendocrine tumors (pNETs) in a non-invasive manner. METHODS: Patients with pNETs who underwent contrast-enhanced abdominal CT between 2010 and 2022 were included in this retrospective study. Radiomics features were extracted, and five radiomics-based ML models, namely logistic regression (LR), random forest (RF), support vector machine (SVM), XGBoost, and GaussianNB, were developed. The performance of these models was evaluated using a time-independent testing set, and metrics such as sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC) were calculated. The accuracy of the radiomics model was compared to that of needle biopsy. The Shapley Additive Explanation (SHAP) tool and the correlation between radiomics and biological features were employed to explore the interpretability of the model. RESULTS: A total of 122 patients (mean age: 50 ± 14 years; 53 male) were included in the training set, whereas 100 patients (mean age: 48 ± 13 years; 50 male) were included in the testing set. The AUCs for LR, SVM, RF, XGBoost, and GaussianNB were 0.758, 0.742, 0.779, 0.744, and 0.745, respectively, with corresponding accuracies of 73.0%, 70.0%, 77.0%, 71.9%, and 72.9%. The SHAP tool identified two features of the venous phase as the most significant, which showed significant differences among the Ki-67 index or mitotic count subgroups (p < 0.001). CONCLUSIONS: An interpretable radiomics-based RF model can effectively differentiate between G1 and G2/3 of pNETs, demonstrating favorable interpretability. CLINICAL RELEVANCE STATEMENT: The radiomics-based interpretable model developed in this study has significant clinical relevance as it offers a non-invasive method for assessing the pathological grade of pancreatic neuroendocrine tumors and holds promise as an important complementary tool to traditional tissue biopsy. KEY POINTS: • A radiomics-based interpretable model was developed to predict the pathological grade of pNETs and compared with preoperative needle biopsy in terms of accuracy. • The model, based on CT radiomics, demonstrated favorable interpretability. • The radiomics model holds potential as a valuable complementary technique to preoperative needle biopsy; however, it should not be considered a replacement for biopsy.

Learning curve of robotic-assisted splenic vessel-preserving spleen-preserving distal pancreatectomy by one single surgeon: a retrospective cohort study.

AIM: Splenic vessel-preserving spleen-preserving distal pancreatectomy (SVP-SPDP) has a lower risk of splenic infarction than the splenicvessel-sacrificing SPDP, but it is more technically demanding. Learning curve of robotic-assisted SVP-SPDP (RSVP-SPDP) remains unreported. This study sought to analyze the perioperative outcomes and learning curve of RSVP-SPDP by one single surgeon. METHODS: Seventy-four patients who were intended to receive RSVP-SPDP at the First Affiliated Hospital of Sun Yat-sen University between May 2015 and January 2023 were included. The learning curve were retrospectively analyzed by using cumulative sum (CUSUM) analyses. RESULTS: Sixty-two patients underwent RSVP-SPDP (spleen preservation rate: 83.8%). According to CUSUM curve, the operation time (median, 318 vs. 220 min; P < 0.001) and intraoperative blood loss (median, 50 vs. 50 mL; P = 0.012) was improved significantly after 16 cases. Blood transfusion rate (12.5% vs. 3.4%; P = 0.202), postoperative major morbidity rate (6.3% vs. 3.4%; P = 0.524), and postoperative length-of-stay (median, 10 vs. 8 days; P = 0.120) improved after 16 cases but did not reach statistical difference. None of the patients had splenic infarction or abscess postoperatively. CONCLUSION: RSVP-SPDP was a safe and feasible approach for selected patients after learning curve. The improvement of operation time and intraoperative blood loss was achieved after 16 cases.

Development and validation of nomogram to predict lymph node metastasis preoperatively in patients with pancreatic neuroendocrine tumor.

BACKGROUND: Routine lymphadenectomy in pancreatic neuroendocrine tumors (pNETs) is debated. There lacks accurate model to predict lymph node metastasis (LNM) preoperatively in pNETs. Therefore, this study aimed at developing a nomogram in predicting LNM in pNETs preoperatively. METHODS: Patients undergoing surgery from Surveillance, Epidemiology, and End Results (SEER) database (design cohort, n = 2742) and First Affiliated Hospital of Sun Yat-sen University (validation cohort, n = 136) were enrolled. Nomogram was developed based on risk factors determined by logistic regression analyses. The performance of nomogram was evaluated by area under receiver operating characteristics curve (AUC), calibration curve, and decision curve analysis. RESULTS: In design cohort, 915 of 2742 patients had LNM. Tumor in the pancreatic head, T stage, and tumor size were significantly associated with LNM (all p < 0.05). Prediction of nomogram was accurate with AUC of 0.776 in design cohort and 0.622 in validation cohort. The nomogram showed good agreement between prediction and observation in the design and validation cohort. Based on nomogram-predicted risk, patients with higher risk of LNM had worse overall survival over patients with lower risk of LNM (log-rank p < 0.001). CONCLUSIONS: The novel nomogram could accurately predict LNM in pNET preoperatively. For patients with high risk of LNM, lymphadenectomy was recommended.

Robotic-assisted organ-preserving or parenchymal-sparing pancreatectomy in pancreatic benign or low-grade malignant tumors: a single institute's experience.

AIM: Robotic-assisted pancreatectomy has been widely used. Organ-preserving pancreatectomy (OPP) and parenchymal-sparing pancreatectomy (PSP) has been gradually adopted for pancreatic benign or low-grade malignant tumors. This study aimed to evaluate the safety and efficacy of robotic-assisted OPP/PSP in our institute. METHODS: Patients undergoing robotic-assisted OPS/PSP at First Affiliated Hospital of Sun Yat-sen University between July 2015 and October 2021 were included in this study. The short-term and long-term outcomes of patients were retrospectively analyzed. RESULTS: Seventy-two patients were enrolled, including spleen-preserving distal pancreatectomy, central pancreatectomy, duodenum-preserving pancreatic head resection, and enucleation. Patients included were more likely to be young female (female: 46/72, median age: 47 years old). The median intraoperative blood loss and operation time was 50 ml and 255 min, respectively. Clinically relevant postoperative pancreatic fistula was 20.8% (grade B: 15/72, 20.8%; no grade C). The overall complication rate was 22.2% with the median postoperative length-of-stay of 8 days. At a median follow-up time of 28.5 months, the 5-year overall survival and recurrence-free survival rate were 100.0% and 100.0%, respectively. CONCLUSION: The short-term and long-term outcomes of patients receiving robotic-assisted OPP/PSP were acceptable. Robotic-assisted OPP/PSP was a feasible and safe technique for pancreatic benign or low-grade malignant lesions.

Learning curves of resection and reconstruction procedures in robotic-assisted pancreatoduodenectomy by a single surgeon: a retrospective cohort study of 160 consecutive cases.

Background: Robotic-assisted pancreatoduodenectomy (RPD) has been routinely performed in a few of centers worldwide. This study aimed to evaluate the perioperative outcomes and the learning curves of resection and reconstruction procedures in RPD by one single surgeon. Methods: Consecutive patients undergoing RPD by a single surgeon at the First Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) between July 2016 and October 2022 were included. The perioperative outcomes and learning curves were retrospectively analysed by using cumulative sum (CUSUM) analyses. Results: One-hundred and sixty patients were included. According to the CUSUM curve, the times of resection and reconstruction procedures were shortened significantly after 30 cases (median, 284 vs 195 min; P < 0.001) and 45 cases (median, 138 vs 120 min; P < 0.001), respectively. The estimated intraoperative blood loss (median, 100 vs 50 mL; P < 0.001) and the incidence of clinically relevant post-operative pancreatic fistula (29.2% vs 12.5%; P = 0.035) decreased significantly after 20 and 120 cases, respectively. There were no significant differences in the total number of lymph nodes examined, post-operative major complications, or post-operative length-of-stay between the two groups. Conclusions: Optimization of the resection procedure and the acquisition of visual feedback facilitated the performance of RPD. RPD was a safe and feasible procedure in the selected patients.

Values of debulking surgery for unresectable well-differentiated metastatic pancreatic neuroendocrine tumors: a comparative study.

Background and objective: The value of debulking surgery for unresectable well-differentiated metastatic pancreatic neuroendocrine tumor (m-PNET) remains poorly defined. This study aimed to evaluate the outcomes of m-PNET following debulking surgery in our institute. Methods: Patients with well-differentiated m-PNET in our hospital between February 2014 and March 2022 were collected. Clinicopathological and long-term outcomes of patients treated with radical resection, debulking surgery, and conservative therapy were compared retrospectively. Results: Fifty-three patients with well-differentiated m-PNET were reviewed, including 47 patients with unresectable m-PNET (debulking surgery, 25; conservative therapy, 22) and 6 patients with resectable m-PNET (radical resection). Patients undergoing debulking surgery had a post-operative Clavien-Dindo ≥ III complication rate of 16.0% without mortality. The 5-year overall survival (OS) rate of patients treated with debulking surgery was significantly higher than that of those treated with conservative therapy alone (87.5% vs 37.8%, log-rank P = 0.022). Besides, the 5-year OS rate of patients treated with debulking surgery was comparable to that of patients with resectable m-PNET undergoing radical resection (87.5% vs 100%, log-rank P = 0.724). Conclusions: Patients with unresectable well-differentiated m-PNET who underwent resection had better long-term outcomes than those who received conservative therapy alone. The 5-year OS of patients undergoing debulking surgery and radical resection were comparable. Debulking surgery could be considered for patients with unresectable well-differentiated m-PNET if no contraindication exists.

Evaluation of the short-term outcomes of robotic-assisted radical resection for perihilar cholangiocarcinoma: a propensity-scored matching analysis.

Background: The application of robotic-assisted radical resection in perihilar cholangiocarcinoma (pCCA) remains poorly defined. This study aimed to evaluate the safety and efficacy of robotic-assisted radical resection for pCCA in our institute. Methods: Between July 2017 and July 2022, pCCA patients undergoing robotic-assisted and open radical resection at First Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) were included. The short-term outcomes were compared by using propensity-scored matching (PSM) analysis. Results: Eighty-six pCCA patients were enrolled. After PSM at a ratio of 1:2, 10 and 20 patients were assigned to the robotic-assisted and open groups, respectively. There were no significant disparities in the clinicopathological features between the two groups. The robotic-assisted group had significantly longer operation time (median: 548 vs 353 min, P = 0.004) and larger total number of lymph nodes examined (median: 11 vs 5, P = 0.010) than the open group. The robotic-assisted group tended to have a lower intraoperative blood loss (median: 125 vs 350 mL, P = 0.067), blood transfusion rates (30.0% vs 70.0%, P = 0.056), and post-operative overall morbidities (30.0% vs 70.0%, P = 0.056) than the open group, even though the differences were not statistically significant. There were no significant differences in the negative resection margin, post-operative major morbidities, or post-operative length-of-stay between the robotic-assisted and open groups (all P > 0.05). Conclusions: Robotic-assisted radical resection of pCCA may get a larger total number of lymph nodes examined than open surgery. Provided robotic-assisted surgery may be a feasible and safe technique for selected pCCA patients.

Values of spleen-preserving distal pancreatectomy in well-differentiated non-functioning pancreatic neuroendocrine tumors: a comparative study.

Background: The feasibility of spleen-preserving distal pancreatectomy (SPDP) to treat well-differentiated non-functioning pancreatic neuroendocrine tumors (NF-pNETs) located at the body and/or tail of the pancreas remains controversial. Distal pancreatectomy with splenectomy (DPS) has been widely applied in the treatment of NF-pNETs; however, it may increase the post-operative morbidities. This study aimed to evaluate whether SPDP is inferior to DPS in post-operative outcomes and survivals when being used to treat patients with NF-pNETs in our institute. Methods: Clinicopathological features of patients with NF-pNETs who underwent curative SPDP or DPS at the First Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) between January 2010 and January 2022 were collected. Short-term outcomes and 5-year survivals were compared between patients undergoing SPDP and those undergoing DPS. Results: Sixty-three patients (SPDP, 27; DPS, 36) with well-differentiated NF-pNETs were enrolled. All patients had grade 1/2 tumors. After identifying patients with T1-T2 NF-pNETs (SPDP, 27; DPS, 15), there was no disparity between the SPDP and DPS groups except for tumor size (median, 1.4 vs 2.6 cm, P = 0.001). There were no differences in operation time (median, 250 vs 295 min, P = 0.478), intraoperative blood loss (median, 50 vs 100 mL, P = 0.145), post-operative major complications (3.7% vs 13.3%, P = 0.287), clinically relevant post-operative pancreatic fistula (22.2% vs 6.7%, P = 0.390), or post-operative hospital stays (median, 9 vs 9 days, P = 0.750) between the SPDP and DPS groups. Kaplan-Meier curve showed no significant differences in the 5-year overall survival rate (100% vs 100%, log-rank P > 0.999) or recurrence-free survival (100% vs 100%, log-rank P > 0.999) between patients with T1-T2 NF-pNETs undergoing SPDP and those undergoing DPS. Conclusions: In patients with T1-T2 well-differentiated NF-pNETs, SPDP could achieve comparable post-operative outcomes and prognosis compared with DPS.