information fusion

Task 1: Data mining and information fusion for tumor response prediction\\ #

Updated September 6th 2016

This task has started in November, 2012, with the recruitment of a PhD student, Mohamed Majdoub. M. Majdoub has started working on the processing and analysis of PET/CT images to derive new shape and heterogeneity metrics (also called radiomics) for characterization of functional tumor volumes that would be able to offer some predictive value on prognosis or response to therapy.

Beside technical investigations on these new features (robustness, correlation with volume, comparison of different calculation methods), he also investigated their potential predictive value in various cohorts of patients with breast and lung cancer (using the FDG radiotracer for metabolism) and Head and Neck cancer (with the FLT radiotracer for cellular proliferation).

His goal was to extract pertinent features from images within the context of response to (radio)chemotherapy prediction (breast, H&N), as well as prognosis of recurrence-free and overall survival (H&N, lung).

In addition to the image processing and analysis part of his work, M.Majdoub has also investigated the use of classification tools such as logistic regression, random forest algorithms, and support vector machines in order to incorporate multiple image-derived and other contextual variables (age, sex, histology, etc.) in a predictive model.

In order to extract image-derived features (radiomics) relevant for prognosis or response to therapy from PET images, M. Majdoub has exploited a pipeline of image processing tools such as noise filtering, partial volume effects correction, metabolic tumor volume automatic delineation, 3D geometrical shape metrics, and heterogeneity quantification through textural features analysis.

Results from the technical investigations suggested that new calculations of heterogeneity metrics could provide valuable prognostic value in addition to standard clinical variables or usual PET image-derived metrics such as SUVs or metabolic volume. This was demonstrated in a multi-centric cohort of 555 patients with 5 different cancer types (lung, esophageal, breast, head and neck, cervix), with prognostic value results on 112 esophageal cancer patients and 101 lung cancer patients. These results were published in the J. Nucl Med [1]. This study has been cited more than 50 times already according to google scholar and web of science has classified it as a “highly cited paper” (25 citations, top 1% of its field).

Another technical investigation consisted in studying the variations of textural features in PET images according to various modes of reconstruction that are gaining interest and clinical use, namely reconstructions using point spread function (PSF) modeling to enhance contrast and standardized uptake values recovery in reconstructed images. However within the context of standardization across centers, the use of these new reconstruction modes raises the question of whether these new image-derived features are impacted or not by the established procedure (to apply a filtering to PSF reconstructed images). These results were published in the Eur J Nucl Med Mol Imaging [2]

On the methodological side, we alo wished to further investigate the repeatability of features extracted from PET/CT images. Although numerous studies had already reported on this issue before, including our own work in 2012, most of these studies included a very limited number of patients (between 8 and 15) recruited in single centers. In addition, all previous studies focused on either FDG PET images or diagnostic/planning CT or contrast-enhanced CT, but never the low-dose CT component of PET/CT. We had the opportunity to work with a larger multi-centric cohort recruited in a prospective way in two clinical trials in the USA (ACRIN), Europe and Asia (Merck), totalling 74 patients with double baseline FDG PET/CT datasets. It took a very long time to obtain access to the entireity of the data, however we were finally able to perform the full analysis in the begining of 2016 and the results are now accepted for publication in the J. Nucl Med [3]. This study is the largest radiomics test-retest repeatability study to date, and the first to report results for both the FDG PET and the associated low-dose CT component of PET/CT acquisitions.

This also allowed a further investigation of the potential complementary prognostic value of PET and associated low-dose CT derived heterogeneity (both FDG radiotracer uptake heterogeneity from PET and tissues density from low-dose CT) in a lung cancer cohort of 116 patients, and demonstrated the feasibility of developing a multimodal PET/CT heterogeneity based prognostic model with high stratification power regarding outcome. Preliminary results for this work were presented at the annual meeting of the AAPM in 2014 [4] where it won an award from the Science council session: the physics of cancer, as well as at the SNMMI annual meeting in 2015 [5] and was recently published as a full paper in the Eur J Nucl Med Mol Imaging journal [6]. Finally, we have started recruiting patients in a prospective manner in order to validate the developed nomogram combining radiomics features from both PET and CT components of FDG PET/CT images using a support vector machine classifier. The preliminary results of this prospective validation (still ongoing) were presented in june of 2016 at the annual meeting of the SNMMI [7] where it won the award for best international abstract.

Another study included 171 patients with breast cancer treated with chemotherapy, with two sequential FDG PET/CT scans. An interesting result was that the optimal PET-derived parameter (either SUVmax, volume or TLG) for the prediction of therapy response using the evolution between the baseline scan and the during chemotherapy scan actually depended on the tumor subgroup (HER2+, ER+/HER2- or triple-negative breast cancer). This work has been accepted in Radiology [8]. In addition, we have investigated the prognostic value of PET-derived parameters for disease-free survival and showed that both SUVmax and TLG have prognostic value. This has published in the J Nuc Med [9]. Finally, we have investigated the relationships between clinical and histopathological factors and heterogeneity features in the same cohort. Unfortunately the heterogeneity metrics did not help in differentiating the various molecular subtypes of breast cancer in this larger cohort. These results were reported in the Eur J Nucl Med Mol Imaging [10].

Another original and novel result was obtained on the cohort of H&N cancer patients treated with (radio)chemotherapy that underwent one FLT PET scan before treatment and a second scan during treatment. It was found that the tri-dimensional shape and the intra-tumor radiotracer spatial distribution (heterogeneity) of the proliferative tumor volume (as quantified on FLT PET images) was significantly associated with recurrence-free survival. Tumors’proliferative volumes with more complex 3D shapes and higher tracer uptake heterogeneity led to more and earlier recurrence after treatment that those with less complex shapes and more homogeneous uptake. In addition, the decrease of the complexity of the 3D shape and the heterogeneity of the tumors’ proliferative volumes during treatment was also associated with less or later recurrence. These new features of tumors extracted from PET images were found to be prognostic factors of overall and recurrence-free survival, with higher discriminative power that standard PET measurements such as functional volume and standardize uptake maximum or mean values.

One hypothesis that can explain these results is the fact that the more complex and heterogeneous a tumor volume is (at least according to the cellular proliferation), the more difficult it is to treat efficiently with a homogeneous radiotherapy dose (which is the standard practice and was the case for this cohort of patients) and/or chemotherapy because the drug cannot be efficiently distributed within the tumor. A quasi-spherical shape with homogeneous uptake distribution is easier to cover with adequate dose with radiotherapy, and the drug (chemotherapy) can be distributed within the entire tumor more efficiently.

These results have been presented in an oral presentation at the nuclear medicine meeting in Vancouver (2013) [11], and a complete paper has been written and submitted to various journals, however the small number of patients have prevented us to publish this yet (at least one of the reviewers has consistently rejected the paper on this sole issue).

Most of the results obtained in this task were also incorporated by M. Hatt in a large review that has been published in the Eur J Nucl Med Mol Imaging [12], as well as in various invited talks in institutions or conferences, and workshops and continuous education sessions (see list below:)

Invited talks:

Building prognostic and predictive models based on PET/CT and PET/MR shape and heterogeneity metrics, Society of nuclear medicine annual meeting, continuous education session “Towards Enhanced Prognostic Utility of PET Imaging”, San Diego, USA, Juin 2016.
PET/CT image processing and analysis for quantitative metrics extraction and clinical applications, Seminaire au CHU de Caen et unité INSERM UMR 1199 BioTICLA, Caen, France, Mai 2016.
PET/CT heterogeneity quantification through texture analysis: potential role for prognostic and predictive models, European SocieTy for Radiotherapy & Oncology annual meeting, symposium on “Radiomics – the future of radiotherapy”, Turin, Italie, Mai 2016.
Peut-on améliorer la prise en charge des patients atteints de cancers grâce à une meilleure exploitation des images médicales?, Conférences IBSAM, Brest, France, Avril 2016
Segmentation et caractérisation des tumeurs en imagerie TEP/TDM : 10 ans d’apports des méthodes statistiques, Journée GDR sur l’apprentissage statistique et applications biomédicales, Paris, France, Décembre 2015
Tumor treatment response assessment using PET/CT, American Society for Radiation Oncology annual meeting, educational session “Imaging for treatment response assessment”, San Antonio, USA, Octobre 2015.
Traitement et analyse d’images TEP/TDM pour l’oncologie et la radiothérapie : méthodes, résultats, perspectives,Séminaire du laboratoire iCube, Strasbourg, France, Mai 2015
Characterization of tumors using quantitative features from PET/CT imaging: methods, results, challenges, Séminaire à l’University College of London, London, UK, Avril 2015
Caractérisation morpho-fonctionnelle des tumeurs en oncologie à partir d’imagerie multimodale, Séminaire du Canceropole Grand Est, Strasbourg, France, Mars 2015
Multimodality image derived tumor characterization, European molecular imaging meeting, educational session “Multi-modality and hybrid imaging equipment”, Tübingen, Allemagne, Mars 2015
From multispectral images of nebulae to multimodal PET/CT images of tumors: a young researcher’s journey, Bruce Hasegawa Young Investigator Medical Imaging Science Award acceptance speech – IEEE NSS-MIC, Seattle, USA, Novembre 2014
Quantification et caractérisation des tumeurs en imagerie TEP : défis et solutions, Séminaire mensuel de l’UMR 6301 CNRS ISTCT, Caen, France, Février 2014.
Tumor functional characterization and quantification, 9th International Conference on Dose, Time and Fractionation in Radiation Oncology – From Hyper to Hypo Fractionation in Radiation Oncology, Madison, USA, Septembre 2013.
PET/CT image derived features for response to therapy monitoring: review and perspectives, American Association of Physicists in Medicine annual meeting, imaging symposium “Multi modal PET/CT imaging for therapy response early prediction and therapy monitoring”, Indianapolis, USA, Août 2013.

Workshops and courses:

Perspectives pour l’approche “Radiomics” en TEP/TDM, Formation “Tomographie par Emission de Positon dans le cadre de la radiothérapie guidée par l’image”, Lyon, France, Avril 2016
Actualités sur les méthodes de segmentation d’images TEP, Formation “Tomographie par Emission de Positon dans le cadre de la radiothérapie guidée par l’image”, Lyon, France, Mars 2016
10 ans de développements d’approches de segmentation d’images TEP/TDM : le chemin parcouru, ce qu’il reste à explorer, Workshop SFPM Le traitement d’images en Physique Médicale, Les Sables-d’Olonne, France, Octobre 2015
« Radiomics » et analyse de textures en imagerie TEP/TDM : potentiel et limites, Workshop SFPM Le traitement d’images en Physique Médicale, Les Sables-d’Olonne, France, Octobre 2015
PET/CT/MRI characterization of tumors using textural features: recent results, Visiting professorship, Imaging Research Laboratory, University of Washington, Seattle, USA, Septembre 2015
Texture analysis in PET/CT: the past, the present… any future?, Visiting professorship, Radiology and radiation oncology departments, University of Washington, Seattle, USA, Septembre 2015

References\\ #

1. M. Hatt, M. Majdoub, M. Vallières, F. Tixier, C. Cheze Le Rest, D.Groheux, E.Hindié, A. Martineau, O. Pradier, R. Hustinx, R. Perdrisot, R.Guillevin, I. El Naqa, D. Visvikis. FDG PET uptake characterization through texture analysis: investigating the complementary nature of heterogeneity and functional tumor volume in a multi-cancer site patient cohort. J Nucl Med 2015 56(1):38-44.

2. C. Lasnon, M. Majdoub, B. Lavigne, P. Do, J. Madelaine, D. Visvikis, N. Aide, M. Hatt. 18F-FDG PET/CT heterogeneity quantification through textural features in the era of harmonisation programs: a focus on lung cancer. Eur J Nucl Med Mol Imaging 2016 (in press)

3. M-C.Desseroit, F. Tixier, W. A.Weber, B. A. Siegel, C. Cheze LeRest, D.Visvikis, M.Hatt. Reliability of PET/CT shape and heterogeneity features infunctional and morphological components of Non-Small Cell Lung Cancertumors: a repeatability analysis in a prospective multi-center cohort. J Nucl Med 2016 (in press)

4. M-C. Desseroit, D. Visvikis, F. Tixier, M. Majdoub, R. Perdrisot, R. Guillevin, O. Pradier, C. Cheze Le Rest, M. Hatt. Complementary Prognostic Value of CT and 18F-FDG PET Non-Small Cell Lung Cancer Tumor Heterogeneity Features Quantified Trough Texture Analysis. AAMP annual meeting 2014.

5. M-C.Desseroit, M. Majdoub, F. Tixier, D. Visvikis, C. Cheze-Le Rest, M. Hatt, Quantification of tumor heterogeneity using textural features analysis of PET and CT images provides complementary prognostic value in stage I-III NSCLC, Society of nuclear medicine and molecular imaging annual meeting 2015

6. M-C. Desseroit, D. Visvikis, F. Tixier, M. Majdoub, R. Perdrisot, R. Guillevin, O. Pradier, M. Hatt, C. Cheze Le Rest. Development of a nomogram combining clinical staging with 18F-FDG PET/CT image features in Non-Small Cell Lung Cancer stage I-III. Eur J Nucl Med Mol Imaging 2016; 43(8):1477-1485

7. M-C. Desseroit, F. Tixier, R. Perdrisot, R. Guillevin, D. Visvikis, C. Cheze Le Rest,  M. Hatt, Nomogram for NSCLC exploiting clinical staging, tumor volume and PET/CT heterogeneity features: development using support vector machines in a retrospective cohort and first validation results in prospectively recruited patientsSociety of nuclear medicine and molecular imaging annual meeting 2016

8. D.Groheux, M. Majdoub, A. Martineau, D. Visvikis, M. Espié, M.Hatt, E. Hindié. Early metabolic response to neoadjuvant treatment: 18FDG-PET/CT criteria according to breast cancer subtype. Radiology 2015;277(2):358-71.

9. D.Groheux, M. Majdoub, A. Martineau, D. Visvikis, M. Espié, M.Hatt, E. Hindié. 18FDG uptake and total lesion glycolysis measured at baseline and after 2 courses of neoadjuvant chemotherapy are powerful tools to predict relapse in patients with ER+/HER2- breast cancer. J Nucl Med 2015;56(6):824-831

10. D. Groheux, M. Majdoub, F. Tixier, C. Cheze Le Rest, A. Martineau, P. Merlet, M. Espié, A. de Roquancourt, E. Hindié, M. Hatt. Do clinical, histological or immunohistochemical primary tumor characteristics translate into different 18FDG-PET/CT image features in stage II-III breast cancer? Eur J Nucl Med Mol Imaging 2015;42(11):1682-91

11. Majdoub M, Visvikis D, Tixier F, B. Hoeben, E. Visser, Cheze-LeRest C, Hatt M. Proliferative 18F-FLT PET tumorvolumes characterization for prediction of locoregional recurrenceand disease-free survival in head and neck cancer. Society of nuclear medicine and molecular imaging annual meeting 2013.

12. M. Hatt, F. Tixier, L. Pierce, P.Kinahan, C. Cheze-Le Rest, D. Visvikis. Characterization of PET/CT images using texture analysis: the past, the present… any future? Eur J Nucl Med Mol Imaging 2016 (in press)

Comments are closed.