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 session 07

340P - Prediction of relapse in colon cancer patients by machine learning models combining radiomics and deep features extracted from baseline computed tomography


10 Sep 2022


Poster session 07


Radiological Imaging;  Digital Health and Real World Data (eHealh, Telehealth, Big Data)

Tumour Site

Colon and Rectal Cancer


América Bueno Gómez


Annals of Oncology (2022) 33 (suppl_7): S136-S196. 10.1016/annonc/annonc1048


A. Bueno Gómez1, N. Tarazona Llavero1, A. Alcolado jaramillo2, F. Gimeno-Valiente3, J.A. Carbonell-Asins4, M. Huerta1, T.C. Fleitas1, S. Roselló Keränen1, L.M. Candia5, D. Roda Perez1, P. Moreno Ruiz6, A. Jiménez Pastor6, A. Alberich-Bayarri6, A. Cervantes1

Author affiliations

  • 1 Department Of Medical Oncology, INCLIVA Biomedical Research Institute, 46010 - Valencia/ES
  • 2 Radiodiagnosis, Hospital Clinico Universitario de Valencia, 46010 - Valencia/ES
  • 3 Cancer Evolution And Genome Instability Laboratory, University College London Cancer Institute, London/GB
  • 4 Precision Medicine Unit, Microbiology Service, Fundación INCLIVA, Hospital Clínico Universitario, 46010 - Valencia/ES
  • 5 Oncology Resident, Hospital Clinico Universitario de Valencia, 46010 - Valencia/ES
  • 6 Quibim, Quibim, Quantitative Imaging Biomarkers in Medicine, 46021 - Valencia/ES

Abstract 340P


Up to 40% of localized colon cancer (CC) patients relapse despite optimal initial treatment. Circulating tumor DNA has emerged as a new prognostic biomarker in this setting. However, liquid biopsy tests provide a limited accuracy in the early prediction of relapse. On this basis, radiomics and artificial intelligence (AI) are posed to provide relevant insights in the prediction of relapse and patient outcome. Here we propose a predictive model of CC recurrence using clinical data in combination with radiomics and deep features extracted from baseline computed tomography (CT) images.


A single-center retrospective observational study was designed. Real-world clinical data and CT examinations were collected from 2015 to 2017. Manual segmentation of the tumor was performed slice by slice by a radiologist with 10+ years of experience in CC using ITK-SNAP. Feature extraction techniques were applied to the tumor voxels through an analysis pipeline in Python adapted from Quibim Precision® software (Quibim, Valencia, ES). Feature reduction techniques were applied to select the characteristics providing independent information. The types of features extracted were: Radiomics, Deep features and fractal dimension. Several classifiers were trained with the clinical and image features as the input, to predict relapse from CC. Relapse and non-relapse classes were balanced to the number of slices containing tumor. Train-validation data ratio was 70:30.


Baseline CT exams from 60 localized CC patients were included. 48% and 20% of them had stage III and high-risk stage II, respectively. The remaining patients were diagnosed with low-risk CC. 36.7% of patients relapsed. Merging the clinical data with the radiomics, fractal and deep features provided a high and increased accuracy 95% (CI 95%: 85 – 100) when compared to clinical variables alone (55%, CI 95%: 32 – 78). Random Forest and cross validation model provided the best performance.


The extraction of radiomics and deep features from CT exams in CC patients and their combination in an AI model provide promising insights towards the early prediction of relapse, leading to identify new and accurate prognostic imaging panels.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

A. Cervantes.




N. Tarazona Llavero: Financial Interests, Personal, Advisory Board: Merck; Financial Interests, Personal, Invited Speaker: MERCK, Pfizer; Financial Interests, Institutional, Funding: Natera Inc; Non-Financial Interests, Member: SEOM Committee. T.C. Fleitas: Financial Interests, Personal, Invited Speaker, Update on the ongoing treatment strategies for GEA: Servier; Financial Interests, Personal, Invited Speaker, The clinical impact of NRTK strategies in GI tumors: Bayer. S. Roselló Keränen: Financial Interests, Personal, Invited Speaker: Amgen, MSD, Servier; Financial Interests, Personal, Advisory Board: Amgen, Pierre Fabre, Servier, Sirtex; Financial Interests, Institutional, Other, Local PI: Ability Pharmaceuticals, Astellas, G1 Therapeutics and Ability Pharmaceuticals, Hutchinson, Menarini, Mirati, Novartis, Pfizer, Pierre Fabre, Roche, Seagen. A. Alberich-Bayarri: Financial Interests, Institutional, Stocks/Shares: Quibim. A. Cervantes: Financial Interests, Institutional, Advisory Board: Merck Serono, Amgen, Roche, Transgene, AnHeart Therapeutics; Financial Interests, Institutional, Invited Speaker: Amgen, Roche, Merck Serono, Foundation Medicine; Financial Interests, Personal, Other, Associate Editor: Annals of Oncology, ESMO Open; Financial Interests, Personal, Other, Editor: Cancer Treatment Reviews; Financial Interests, Institutional, Research Grant, Principal Investigator: Actuate Therapeutic, Amgen, Astellas Pharma, Beigene, Bayer, AstraZeneca, BMS, Amcure, FibroGen, Lilly, Genentech, MedImmune, Merck Serono, Novartis, Natera, MSD, Servier, Sierra Oncology, Adaptimmune, Takeda; Non-Financial Interests, Other, General and Scientific Director: INCLIVA Biomedical Research Institute. All other 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.