Objective:To explore the clinical application value of artificial intelligence models in identifying bleeding events and hemorrhagic points during laparoscopic pancreatic surgery.Methods:This single-center retrospective cohort study collected surgical videos of 25 patients undergoing laparoscopic pancreatic surgery at the Department of General Surgery, Peking Union Medical College Hospital from January 2022 to December 2024. Videos within 5 seconds before and after representative bleeding events were captured at 30 frames/s, with 11 666 hemorrhagic-related video frames annotated. Two algorithm models were developed: a pigment-based model and a pigment+optical flow-based model for classification and target recognition of bleeding frames. The training and test sets for the pigment-based algorithm contained 4 692 hemorrhagic and 4 309 non-hemorrhagic frames, while those for the pigment+optical flow model included 1 339 hemorrhagic and 1 326 non-hemorrhagic frames. Performance evaluation was conducted using overlap thresholds, with accuracy and recall rates as key metrics.Results:The pigment-based model achieved 93.8% accuracy (134/143) and 43.3% recall (134/310) in hemorrhagic frame classification. At an overlap threshold of 0.3, the pigment-based model showed 84.1% accuracy (433/515) and 85.4% recall (433/507) in target recognition. When the threshold was increased to 0.5, the pigment+optical flow model demonstrated 88.1% accuracy (354/402) and 89.2% recall (354/397) in hemorrhagic target recognition.Conclusions:It is difficult to distinguish active bleeding from old bleeding completely by pigment information alone. The spatio-temporal features can be effectively extracted by combining pigment and optical flow information, and the bleeding can be accurately identified and located, which has potential clinical application value.

Objective:To explore the clinical application value of artificial intelligence models in identifying bleeding events and hemorrhagic points during laparoscopic pancreatic surgery.Methods:This single-center retrospective cohort study collected surgical videos of 25 patients undergoing laparoscopic pancreatic surgery at the Department of General Surgery, Peking Union Medical College Hospital from January 2022 to December 2024. Videos within 5 seconds before and after representative bleeding events were captured at 30 frames/s, with 11 666 hemorrhagic-related video frames annotated. Two algorithm models were developed: a pigment-based model and a pigment+optical flow-based model for classification and target recognition of bleeding frames. The training and test sets for the pigment-based algorithm contained 4 692 hemorrhagic and 4 309 non-hemorrhagic frames, while those for the pigment+optical flow model included 1 339 hemorrhagic and 1 326 non-hemorrhagic frames. Performance evaluation was conducted using overlap thresholds, with accuracy and recall rates as key metrics.Results:The pigment-based model achieved 93.8% accuracy (134/143) and 43.3% recall (134/310) in hemorrhagic frame classification. At an overlap threshold of 0.3, the pigment-based model showed 84.1% accuracy (433/515) and 85.4% recall (433/507) in target recognition. When the threshold was increased to 0.5, the pigment+optical flow model demonstrated 88.1% accuracy (354/402) and 89.2% recall (354/397) in hemorrhagic target recognition.Conclusions:It is difficult to distinguish active bleeding from old bleeding completely by pigment information alone. The spatio-temporal features can be effectively extracted by combining pigment and optical flow information, and the bleeding can be accurately identified and located, which has potential clinical application value.

ObjectiveTo explore the molecular mechanism implicated in the treatment of primary myelofibrosis （PMF） using Chinese medicinal herbs （CMH） by bioinformatics and molecular dynamics. MethodsData mining was performed to find the high-frequency CMH in treating PMF between the year of 1985 and 2024 by searching CNKI， Chinese Science and Technology Journal Database （CCD）， and China Academic Journal Database （CSPD）. TCMSP， SwissTargetPrediction and related reports were used to collect the main active ingredients of high-frequency CMH and their targets. The PMF datasets GSE44426 and GSE124281 were downloaded from GEO database， and R software was used for data normalization and differentially expressed genes （DEGs） screening. Key module hub genes were obtained by weighted gene co-expression network analysis （WGCNA） analysis. The common intersection genes of active ingredient targets， DEGs and key module hub genes of CMH were selected， and the target network was generated using Cytoscape 3.9.2 software. The core target network was generated by topological analysis， while key pathways were selected by GO and KEGG pathway enrichment analysis， and protein interaction relationships were obtained from the String database， so as to construct drug-ingredient-target network and protein interaction network （PPI） relationship diagrams. Discovery Studio 2020 software was used to perform molecular docking， and the GROMACS program was used to perform molecular dynamics simulation. ResultsA total of 21 prescriptions were collected involving 121 herbs. There were 9 herbs with a frequency ≥10 times， which were Danshen （Radix et Rhizoma Salviae Miltiorrhizae）， Huangqi （Radix Astragali）， Baizhu （Rhizoma Atractylodis Macrocephalae）， Danggui （Radix Angelicae Sinensis）， Dangshen （Radix Codonopsis）， Gancao （Radix et Rhizoma Glycyrrhizae）， Baishao （Radix Paeoniae Alba）， Fuling （Poria） and Shudihuang （Radix Rehmanniae Praeparata） from high- to low-frequency. A total of 98 active ingredients and 1125 potential targets were obtained from 9 high-frequency CMH. GSE44426 and GSE124281 data sets screened out 24 gene samples， including 14 of the healthy control group and 10 of the PMF group， and identified 319 DEGs between the two groups， including 122 up-regulated genes and 197 down-regulated genes. WGCNA screened out 24 co-expression module genes and found that the five modules closely related to the onset of PMF were MEpink， MEdarkred， MEblack， MEgrey， and MEturquoise， involving 7112 key module hub genes. The GO and KEGG enrichment analyses indicated that lipids and the atherosclerosis pathways were mainly involved in the mechanism of above high-frequency CMH in treating PMF， which included six hub protein targets： HSP90AA1， HSP90AB1， SRC， MAPK1， IL1B and IL10. From the drug-ingredient-target network， seven active ingredients of CMH targeting at these six hub targets were found， including verbascoside， verbascos isoflavone， kaempferol， luteolin， naringenin， quercetin and pachymic acid. The molecular docking and molecular dynamics analyses showed that the key CMH were Shudihuang， Huangqi， Baishao， Danshen， Gancao and Fuling， and among the seven active ingredients， calycosin had the highest binding affinity with HSP90AB1. ConclusionThe main CMH for the treatment of PMF may be Shudihuang， Huangqi， Baishao， Danshen， Gancao and Fuling， and the active ingredients include verbascoside， verbascos isoflavones， kaempferol， luteolin， naringenin， quercetin and pachymic acid. The relevant targets are HSP90AA1， HSP90AB1， SRC， MAPK1， IL-10， and IL-1β， and the most critical pathways are lipid and atherosclerosis pathways.

Objective:To introduce a new TIANJI robot assisted targeted puncture technique, and discuss the feasibility and clinical effect of transforaminal percutaneous endoscopic lumbar discectomy (tPELD) using this technique.Methods:The first 14 consecutive cases of single level lumbar disc herniation who underwent robot assisted tPELD procedure were retrospectively analyzed. The mean age was 46.3±16.0 years old (ranged from 16-72). After data transferred from C-arm to robot system and automatic registration, surgeons made plans of the trajectory on robot system based on intraoperative 3-dimensional images of lumbar spine. Move robotic arm to planned position, guide an accurate puncture pathway and establish working cannula. 25 consecutive patients who underwent conventional C-arm assisted tPELD surgery during the same period of time were assessed as the controlled group. The mean age was 45.5±13.7 years old (ranged from 16-68). All patients were followed up for 12 months. Clinical effect was assessed by visual analogue scale (VAS), Oswestry disability index (ODI) and Modified Macnab criteria. Intraoperative parameters and surgery-related complications were recorded.Results:The baseline data of age, surgical level, types of herniation, preoperative VAS scores and ODI had no significant difference between two groups ( P>0.05). In robot group, one case was converted to open microdiscectomy during operation due to technical failure. The other thirteen cases had successful robot assisted tPELD surgeries and were assessed accordingly. The new technique had good clinical outcomes. The immediate post-operative VAS score 2.85±1.79 and the last follow-up VAS score 1.50±1.04 were both significantly decreased than that before surgery 7.62±0.92 ( F=69.747, P<0.01); the last follow-up ODI 18.89%±12.16% was significantly reduced from the pre-operative ODI 71.19%±12.12% ( t=15.430, P<0.01). Between two groups, the immediate post-operative VAS score ( t=0.568, P=0.574), the last follow-up VAS score ( t=0.713, P=0.481), and last follow-up ODI had no significant difference ( t=0.171, P=0.865). The excellent or good rate of modified Macnab criteria at the last follow-up was 92.30% in robot group, comparing to 84.0% in controlled group. The fluoroscopic times during surgery of robot group 8.8±5.5 was significantly lowered the in controlled group 21.3±8.3 ( P<0.01). One case in robot group and two cases in controlled group had recurrence during follow-up period (recurrence rate 7.7% vs. 8.3%). However, there was no significant complications such as nerve root injury, dura injury or increased intracranial pressure in both groups. Conclusion:This study confirmed the feasibility of this new technique. Preliminary results indicated that TIANJI robot could help to build an easy, accurate and safe procedure of tPELD surgery.