ESTRO 2024 - Abstract Book
S4412
Physics - Machine learning models and clinical applications
ESTRO 2024
Results:
Compared with variant models that input only intensity or only morphological information from CT, combining both as the CNN model input improved the diagnostic accuracy. The analysis revealed that the morphology information was crucial for discriminating SBD from CBD and enabled a relatively high accuracy. However, high accuracy could not be achieved for CBD when only morphological information is available, the combination of the two achieved optimal results. For the reader study, the proposed CNN model attained a significantly higher accuracy than hepatobiliary surgeons for CBD. Two differently trained CNN models (different data category configurations) achieved sensitivities of 0.817 and 0.950, whereas two surgeons achieved 0.433 and 0.333. In terms of specificity, the two models achieved 0.911 and 0.678, compared with 0.867 and 1.0 for the two surgeons. The AUC values of the models were 0.897 and 0.884. In each category, the accuracies of the proposed two differently trained CNN models were significantly higher than those of the surgeons for CBD, but slightly lower for SBD. For each category, the accuracies achieved by the proposed CNN model were 0.800, 0.833, 1.000, 0.800, and 0.933 for categories A–E, respectively.
Conclusion:
Our proposed model provides an objective and accurate diagnostic approach to better diagnose CBD, which explores dilated biliary tract differences through multiple variants models and training strategies that distinguish biliary features of SBD from those of CBD. While surgeons tend to misdiagnose CBD as SBD, the proposed method offers an increased ability to recognize CBD. Thus, the model demonstrates the potential to assist radiologists and hepatobiliary surgeons in differentiating between CBD and SBD and making more appropriate treatment decisions. Thus, it is a valuable tool for clinical diagnosis.
Keywords: Deep learning, Congenital biliary dilatation
References:
[1] Kumar S, Singla S, Kumar S, Singh RK, Raj AK, Mandal M. Tactics, techniques, and challenges in the management of giant choledochal cyst in adolescents and adults: a retrospective cohort study. Langenbecks Arch Surg 2021;406:1925-1933. PMID:34021416. doi: 10.1007/s00423-021-02209-6. [2] Yuan H, Dong G, Zhang N, Sun X, Zhao H. Minimally invasive strategy for type I choledochal cyst in adult: combination of laparoscopy and choledochoscopy. Surg Endosc 2021;35:1093-1100. PMID:32095950. doi: 10.1007/s00464-020-07473-z.
[3] Mabrut JY, Bozio G, Hubert C, Gigot JF. Management of congenital bile duct cysts. Dig Surg 2010;27:12-18. PMID:20357446. doi: 10.1159/000268109.
[4] Gomes C, Tivnan P, McAneny D, Tseng JF, Tkacz J, Sachs TE. Choledochal Cyst or Benign Biliary Dilation: Is Resection Always Necessary? J Gastrointest Surg 2021;25:2353-2357. PMID:33483915. doi: 10.1007/s11605-020-04896-w.
[5] Aoki Y, Aida J, Kawano Y, Nakamura KI, Izumiyama-Shimomura N, Ishikawa N, et al. Telomere length of gallbladder epithelium is shortened in patients with congenital biliary dilatation: measurement by quantitative fluorescence in situ hybridization. J Gastroenterol 2018;53:291-301. PMID:29143121. doi: 10.1007/s00535-017-1411-5.
Made with FlippingBook - Online Brochure Maker