Offre de Post-Doc au laboratoire LITIS à Rouen

With the increasing amount and complexity of available medical data, there is a real and growing need for the development of systems based on models for predicting the results of a therapeutic treatment to assist physicians in decision-making. In the field of radiological oncology, particularly in the case of individualized medicine, these models combine both predictive and prognostic data from clinical and molecular data, medical imaging to accurately predict tumor response to initial treatment and patient follow-up rate [1]. This allows to customize the treatment according to the prediction.

Radiomics is a new and evolving field within the medical imaging community. It consists of 1) automatic extraction of characteristics from medical images such as PET-CT or MRI and 2) construction of predication models based on extracted characteristics. Some advanced methods in radiomics have been proposed. For example, to predict treatment outcomes in patients with cervical cancer or head and neck cancer, El Naqa et al. [2] developed a predictive model based on logistic regression using the shape and texture characteristics extracted from PET images. In [3], temporal changes in FDG-PET characteristics were adopted to predict the response of esophageal cancer to chemoradiotherapy using the Support Vector Machine (SVM) method. In [4], a logistic regression model based on the joint quantification of FDG-PET and MRI textures was developed for the prediction of pulmonary metastases in soft tissue sarcomas of the extremities. Our recent work with the Quantif-LITIS team selects features using SVM [5], Random Forest [6] and belief functions [7] for the prediction of pulmonary and esophageal tumor recurrence after therapeutic treatment.  Although radiomic analysis has been claimed to be useful in predicting the results of cancer treatment, its solid application is still hampered by multiple difficulties: the low quality of PET / CT images, the large variability in sizes and forms of tumor, no-prior knowledge on the relevance of the characteristics for the prediction, etc.

In this project we propose to develop new models of prediction based on the research in the field of non-local model theory  [8]. In our previous work on radiomics, image characteristics were studied. Based on the extraction of features that we have proposed, the project will focus on the construction of prediction models. The characteristics will be represented by the graph in a very large space in which the decision rules will be studied and established. We are conducting the project in close collaboration with the Center of Cancer Research – Henri Becquerel Center in Rouen, which will provide not only medical data but also medical and clinical expertise.


  1. Philippe Lambin, Ruud G. P. M. van Stiphout, Maud H. W. Starmans, Emmanuel Rios-Velazquez, Georgi Nalbantov, al. Predicting outcomes in radiation oncology—multifactorial decision support systems, Nature Reviews Clinical Oncology 10, 27-40 (January 2013).
  2. El Naqa, P. Grigsby, A. Apte, E. Kidd, E. Donnelly, D. Khullar, S. Chaudhari, D. Yang, M. Schmitt, R. Laforest, et al., “Exploring feature-based approaches in PET images for predicting cancer treatment outcomes,” Pattern Recognition, vol. 42, no. 6, pp. 1162–1171, 2009.
  3. Zhang, S. Tan, W. Chen, S. Kligerman, G. Kim, W. D. D’Souza, M. Suntharalingam, and W. Lu, “Modeling pathologic response of esophageal cancer to chemoradiation therapy using spatial-temporal 18 F-FDG PET features, clinical parameters, and demographics,” International Journal of Radiation Oncology* Biology* Physics, vol. 88, no. 1, pp. 195–203, 2014.
  4. Vallières, C. Freeman, S. Skamene, I. El Naqa, et al., “A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities,” Physics in Medicine and Biology, vol. 60, no. 14, p. 5471, 2015.
  5. Hongmei Mi, Caroline Petitjean, Bernard Dubray, Pierre Vera, Su Ruan, « Robust Feature Selection to Predict Tumor Treatment Outcome », Elsevier,Artificial Intelligence in Medicine, Volume 64, Issue 3,  Pages 195–204, July 2015.
  6. Desbordes Paul, Ruan Su, Modzelewski Romain, Vauclin Sébastien, Vera Pierrea, Gardin Isabelle, Feature selection for outcome prediction in oesophageal cancer using genetic algorithm and random forest classifier, Elsavier Computerized Medical Imaging and Graphics, In Press,
  7. Chandrashekar and F. Sahin, “A survey on feature selection methods,” Computers & Electrical Engineering, vol. 40, no. 1, pp. 16–28, 2014.
  8. Elmoataz, X. Desquesnes, Z. Lakhdari, O. Lezoray, Nonlocal infinity Laplacian equation on graphs with applications in image processing and machine learning, Mathematics and Computers in Simulation, vol. 102, pp. 153–163, 2014.
  9. Elmoataz, X. Desquesnes, On the game p-Laplacian on Weighted Graphs with Applications in Image Processing and Data Clustering, European Journal of Applied Mathematics, 2017.

Functional area: Rouen, France
Team: Quantif, LITIS laboratory

Duration: 12 months
Start date: from the 1st of March 2018
Net salary: 2100 euros per month

Skills and Profile

  • PhD in computer science, data science, image processing (medical image processing), computer vision or applied maths.
  • Excellent programming skills.


Su Ruan <>, Jérôme Lapuyade-Lahorgue <>