Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster Display session

52P - Prediction of postoperative recurrence in pulmonary NETs using an artificial intelligence model: Multilayer Perceptron

Date

21 Mar 2025

Session

Poster Display session

Presenters

Piero Paravani

Citation

Annals of Oncology (2025) 10 (suppl_3): 1-7. 10.1016/esmoop/esmoop104347

Authors

P. Paravani1, A. La Salvia2, M. Martiradonna1, B. Giorgini1, B. Golisano1, M. Mancini3, A. Siciliani4, V. Zamponi1, R. Mazzilli1, A. Faggiano1

Author affiliations

  • 1 Endocrinology Unit, Department Of Clinical And Molecular Medicine, Sapienza University Of Rome, Sant’andrea University Hospital, Enets Center Of Excellence, Sapienza University of Rome, 00185 - Rome/IT
  • 2 National Center For Drug Research And Evaluation, Istituto Superiore di Sanità (ISS), 161 - Rome/IT
  • 3 Department Of Clinical And Molecular Medicine, Morphologic And Molecular Pathology Unit, St. Andrea University Hospital, Azienda Ospedaliera Sant'Andrea, 00189 - Rome/IT
  • 4 Department Of Thoracic Surgery, Sant'andrea Hospital-sapienza University, Azienda Ospedaliera Sant'Andrea, 00189 - Rome/IT

Resources

This content is available to ESMO members and event participants.

Abstract 52P

Background

Radical surgical treatment is the standard-of-care in localized and locally advanced Lung Neuroendocrine tumors (Lu-NET). Although considered indolent, recurrence rates, especially in the first 2 years post-surgery, are not negligible.

Methods

The aim of this study is to design and evaluate a predictive model of post-surgical recurrence in Lu-NET. Data were collected from all patients referred to our center from 01/2021 undergoing radical surgery with at least 12 months of follow-up. Clinicopathological information was collected and impact selected factors on RFS was evaluated by KM analysis. Three models were constructed using selected predictors of recurrence using (ML) algorithms, specifically a multilayer perceptron (MLP). 70% of the information was used as data for the training group and 30% was used for the verification group. A Repeated Holdout method was used for internal validation. Accuracy and AUC were calculated.

Results

We included 81 patients (54=F); 70 typical carcinoid and 11 atypical carcinoid(AC). Ki-67% was<3% in 53 and mitotic index was<2 in 64 cases. A functional syndrome was present in 6. Stage was I in 57, and 15 had LV1. 13 recurrence events were observed (16.7%), including 5 during the first 12 months, 11 in 24 months. According to KM analysis, age (p<0.001), AC (p=0.006), Ki-67% (p<0.001), IM (p<0.001) functional syndrome (p<0.001), stage (p<0.001) LV1 (p=0.023) were found to be predictors of shorter RFS. Three models were then constructed to predict recurrence at 12m, 24m, and independently of time using the predictors identified at KM analysis. The model at 12m had an AUC=0.911 (SD:0.134) with accuracy=96.49% (SD:3.32%); the predictive model at 24m AUC=0.883 (SD:0.086) and accuracy=91.10% (SD:7.17%). Finally, the time-independent model demonstrated an AUC=0.922 (SD:0.054) and accuracy=88.78% (SD:8.62%).

Conclusions

Valid predictive model based on ML could be an important tool to support clinician. For the first time in this preliminary study, a similar model was investigated in Lu-NETs with promising results. Results must be interpreted considering the study limitations, like its retrospective nature and low numerosity. Subsequent prospective study will be able to confirm these results.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.