pr\u00e9sentent<\/p>\n\n\n\n
Autoris\u00e9 \u00e0 pr\u00e9senter ses travaux en vue de l\u2019obtention du Doctorat de l’Institut Polytechnique de Paris, pr\u00e9par\u00e9 \u00e0 l’Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis en :<\/p>\n\n\n\n
le MERCREDI 4 NOVEMBRE 2026 \u00e0 10h00<\/p>\n\n\n\n
\u00e0<\/p>\n\n\n\n
Amphith\u00e9\u00e2tre 0C06 Amphi 6 hybride
19 Place Marguerite Perey, 91120 Palaiseau<\/p>\n\n\n\n
Membres du jury :<\/strong><\/p>\n\n\n\n M. Rapha\u00ebl FOURNIER-S\u2019NIEHOTTA<\/strong>, Ma\u00eetre de conf\u00e9rences, Sorbonne Universit\u00e9, FRANCE – Examinateur<\/p>\n\n\n\n Mme Sonia GARCIA-SALICETTI<\/strong>, Professeure, Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis, FRANCE – Directrice de th\u00e8se<\/p>\n\n\n\n Mme Nesma HOUMANI<\/strong>, Ma\u00eetresse de conf\u00e9rences, Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis, FRANCE – Co-encadrante de th\u00e8se<\/p>\n\n\n\n R\u00e9sum\u00e9 :<\/strong><\/p>\n\n\n\n Les maladies h\u00e9patiques chroniques \u00e9voluent par la fibrose puis la cirrhose vers le stade terminal, et favorisent le carcinome h\u00e9patocellulaire (CHC). La transplantation h\u00e9patique permet de traiter les deux pathologies, mais les greffons restent rares et doivent \u00eatre r\u00e9partis entre les candidats. La priorisation repose sur des scores pronostiques estimant le risque \u00e0 court terme de sortie de liste pour d\u00e9c\u00e8s ou aggravation. Le MELD et ses \u00e9volutions r\u00e9gissent cette allocation, tandis que les candidats atteints de CHC re\u00e7oivent des points d’exception ou font l\u2019objet de mod\u00e8les sp\u00e9cifiques. D\u00e9riv\u00e9s de r\u00e9gressions de Cox sur peu de variables biologiques, ces scores appliquent les m\u00eames pond\u00e9rations fixes \u00e0 tous les candidats. Ils ne repr\u00e9sentent donc pas les interactions non lin\u00e9aires entre dysfonctions h\u00e9patique, r\u00e9nale et syst\u00e9mique, et ignorent les mesures r\u00e9p\u00e9t\u00e9es du suivi. Ils comportent aussi des in\u00e9quit\u00e9s document\u00e9es, notamment envers les femmes dont la moindre masse musculaire abaisse la cr\u00e9atinine et donc la s\u00e9v\u00e9rit\u00e9 calcul\u00e9e. Les travaux existants en apprentissage automatique (ML) restent souvent peu interpr\u00e9tables, se limitent \u00e0 un horizon unique et mod\u00e9lisent s\u00e9par\u00e9ment la population CHC. Cette th\u00e8se d\u00e9veloppe des m\u00e9thodes de ML explicables et transf\u00e9rables pour la pr\u00e9diction personnalis\u00e9e de la sortie de liste, avec et sans CHC, \u00e0 partir du registre STAR de l’OPTN\/UNOS (2002 – 2023). Nous \u00e9tablissons d’abord les fondements d’analyse de survie du probl\u00e8me. Le suivi en liste est un processus \u00e0 risques concurrents o\u00f9 sortie de liste et transplantation s’excluent mutuellement, et censurer la transplantation en surestime le risque absolu. \u00c0 l’aide de mod\u00e8les de Cox et de Fine-Gray, nous identifions les facteurs de risque de la sortie de liste. Appliqu\u00e9e aux disparit\u00e9s li\u00e9es au sexe, cette analyse montre que le d\u00e9savantage des femmes refl\u00e8te un d\u00e9faut de calibration plut\u00f4t que de discrimination, largement corrig\u00e9 par le MELD 3.0. Nous reformulons ensuite le probl\u00e8me en une classification \u00e0 horizon fixe, permettant l’utilisation de mod\u00e8les de ML flexibles tout en exploitant l\u2019information longitudinale. Nous introduisons le score ELTM, un mod\u00e8le de gradient boosting pr\u00e9disant la sortie de liste \u00e0 3, 6 et 12 mois pour la population g\u00e9n\u00e9rale sans CHC. La s\u00e9lection de variables par SHAP r\u00e9duit le mod\u00e8le \u00e0 un ensemble compact sans perte de performance, et les explications locales r\u00e9v\u00e8lent l’\u00e9volution du risque individuel. Nous \u00e9tendons cette approche au CHC avec ELM-HCC, combinant s\u00e9v\u00e9rit\u00e9 h\u00e9patique et marqueurs de charge tumorale. Au-del\u00e0 des contributions par variable, nous exploitons la repr\u00e9sentation SHAP comme un espace de similarit\u00e9 entre patients, identifiant des profils de risque aux trajectoires de mortalit\u00e9 distinctes. Une contribution majeure concerne le transfert de repr\u00e9sentations pronostiques entre populations. Celles-ci partageant des d\u00e9terminants majeurs de mortalit\u00e9, nous comparons deux strat\u00e9gies de transfert de la large cohorte sans CHC vers la cohorte CHC, plus limit\u00e9e : un cadre neuronal multit\u00e2che apprenant une repr\u00e9sentation partag\u00e9e, et un transfert explicite en deux \u00e9tapes int\u00e9grant ELTM comme m\u00e9ta-variable apprise dans un mod\u00e8le d\u00e9di\u00e9 au CHC. Le transfert explicite donne le mod\u00e8le le plus performant et le moins complexe, montrant qu’une repr\u00e9sentation compacte de la s\u00e9v\u00e9rit\u00e9 h\u00e9patique capture l’essentiel de l’information pronostique commune. Enfin, nous validons ELTM en analyse de survie avec des m\u00e9triques ajust\u00e9es pour la censure, face aux scores MELD et \u00e0 des mod\u00e8les de survie. Entra\u00een\u00e9 \u00e0 horizon fixe, ELTM fournit des estimations coh\u00e9rentes jusqu’\u00e0 l’\u00e9v\u00e9nement. Ces contributions permettent de : (i) analyser les disparit\u00e9s sous risques concurrents, (ii) fournir des estimations personnalis\u00e9es et multi-horizons pour les candidats avec et sans CHC, et (iii) apporter des explications individuelles transparentes, pour une allocation plus \u00e9quitable. The liver holds a large functional reserve that masks damage until disease becomes advanced. Chronic liver disease then progresses through fibrosis and cirrhosis toward end-stage liver disease, and creates the conditions for hepatocellular carcinoma (HCC). Liver transplantation is the definitive treatment for both, yet donor organs remain scarce and must be distributed among competing candidates. Allocation therefore relies on prognostic scores estimating the short-term risk of death or removal from the waiting list, referred to as waitlist dropout. In current practice, MELD and its variants govern this prioritization, while HCC candidates are granted exception points or assessed through tumor-specific models. Derived from Cox regression on few laboratory variables, these scores apply identical fixed weights to all candidates. They therefore cannot capture the nonlinear interactions between hepatic, renal and systemic dysfunction and discard the repeated measurements collected during follow-up. They also carry documented inequities, most notably the disadvantage of women whose lower muscle mass deflates serum creatinine and hence their computed severity. Existing Machine Learning (ML) studies often remain difficult to interpret, focus on a single horizon, and model the HCC population separately. This thesis develops accurate, explainable and transferable ML methods for the personalized prediction of waitlist mortality for candidates with and without HCC. To this end, we use the OPTN\/UNOS STAR registry, covering adult candidates listed between February 2002 and September 2023. We first establish the survival foundation of the problem. Waitlist follow-up is a competing-risks process where dropout and transplantation prevent one another, and treating transplantation as censoring overstates the absolute dropout risk. Using cause-specific Cox and Fine-Gray models, we identify the determinants of dropout. Applied to gender, this framework clarifies the disadvantage of women: it reflects a failure of calibration rather than of ranking, and is largely corrected by the sex-recalibrated MELD 3.0. We then restate this cause-specific target as a horizon-specific classification problem, enabling flexible ML models while exploiting repeated observations. We introduce the ELTM score, a gradient-boosting model predicting dropout within 3, 6 and 12 months for the general non-HCC population, interpreted through SHAP. Feature selection optimizes the model to a compact variable set without loss of performance, while local explanations reveal how individual risk evolves over time. We extend this approach to HCC candidates with ELM-HCC, combining hepatic severity with tumor burden markers. Beyond feature attribution, we exploit the SHAP representation as a patient similarity, identifying risk profiles reflecting distinct mortality pathways. A main contribution concerns cross-population risk-representation transfer. Because both populations share core determinants of mortality, we compare two strategies for transferring knowledge from the large non-HCC cohort to the smaller HCC one: a neural multi-task framework learning a shared representation, and an explicit two-stage transfer in which ELTM enters a dedicated HCC model as a learned meta-feature. The explicit transfer yields the strongest and least complex predictor, showing that a compact representation of liver disease severity captures much of the shared prognostic information. Finally, we validate ELTM within a time-to-event framework using censoring-adjusted metrics, benchmarked against MELD scores and survival models. ELTM, trained at a fixed horizon, is shown to provide consistent time-to-event estimates. Together, these contributions: (i) analyse disparities under competing risks, (ii) deliver personalized and multi-horizon risk estimates for both populations, and (iii) provide transparent, patient-level explanations supporting equitable organ allocation.<\/p>\n","protected":false},"excerpt":{"rendered":" L’Ecole doctorale : Ecole Doctorale de l’Institut Polytechnique de Paris et le Laboratoire de recherche SAMOVAR – Services r\u00e9partis, Architectures, Mod\u00e9lisation, Validation, Administration des R\u00e9seaux pr\u00e9sentent l\u2019AVIS DE SOUTENANCE de Monsieur Abdelghani HALIMI Autoris\u00e9 \u00e0 pr\u00e9senter ses travaux en vue de l\u2019obtention du Doctorat de l’Institut Polytechnique de Paris, pr\u00e9par\u00e9 \u00e0 l’Institut Polytechnique de Paris […]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ocean_front_end_style_editor":"no","ocean_post_layout":"","ocean_both_sidebars_style":"","ocean_both_sidebars_content_width":0,"ocean_both_sidebars_sidebars_width":0,"ocean_sidebar":"","ocean_second_sidebar":"","ocean_disable_margins":"enable","ocean_add_body_class":"","ocean_shortcode_before_top_bar":"","ocean_shortcode_after_top_bar":"","ocean_shortcode_before_header":"","ocean_shortcode_after_header":"","ocean_has_shortcode":"","ocean_shortcode_after_title":"","ocean_shortcode_before_footer_widgets":"","ocean_shortcode_after_footer_widgets":"","ocean_shortcode_before_footer_bottom":"","ocean_shortcode_after_footer_bottom":"","ocean_display_top_bar":"default","ocean_display_header":"default","ocean_header_style":"","ocean_center_header_left_menu":"","ocean_custom_header_template":"","ocean_custom_logo":0,"ocean_custom_retina_logo":0,"ocean_custom_logo_max_width":0,"ocean_custom_logo_tablet_max_width":0,"ocean_custom_logo_mobile_max_width":0,"ocean_custom_logo_max_height":0,"ocean_custom_logo_tablet_max_height":0,"ocean_custom_logo_mobile_max_height":0,"ocean_header_custom_menu":"","ocean_menu_typo_font_family":"","ocean_menu_typo_font_subset":"","ocean_menu_typo_font_size":0,"ocean_menu_typo_font_size_tablet":0,"ocean_menu_typo_font_size_mobile":0,"ocean_menu_typo_font_size_unit":"px","ocean_menu_typo_font_weight":"","ocean_menu_typo_font_weight_tablet":"","ocean_menu_typo_font_weight_mobile":"","ocean_menu_typo_transform":"","ocean_menu_typo_transform_tablet":"","ocean_menu_typo_transform_mobile":"","ocean_menu_typo_line_height":0,"ocean_menu_typo_line_height_tablet":0,"ocean_menu_typo_line_height_mobile":0,"ocean_menu_typo_line_height_unit":"","ocean_menu_typo_spacing":0,"ocean_menu_typo_spacing_tablet":0,"ocean_menu_typo_spacing_mobile":0,"ocean_menu_typo_spacing_unit":"","ocean_menu_link_color":"","ocean_menu_link_color_hover":"","ocean_menu_link_color_active":"","ocean_menu_link_background":"","ocean_menu_link_hover_background":"","ocean_menu_link_active_background":"","ocean_menu_social_links_bg":"","ocean_menu_social_hover_links_bg":"","ocean_menu_social_links_color":"","ocean_menu_social_hover_links_color":"","ocean_disable_title":"default","ocean_disable_heading":"default","ocean_post_title":"","ocean_post_subheading":"","ocean_post_title_style":"","ocean_post_title_background_color":"","ocean_post_title_background":0,"ocean_post_title_bg_image_position":"","ocean_post_title_bg_image_attachment":"","ocean_post_title_bg_image_repeat":"","ocean_post_title_bg_image_size":"","ocean_post_title_height":0,"ocean_post_title_bg_overlay":0.5,"ocean_post_title_bg_overlay_color":"","ocean_disable_breadcrumbs":"default","ocean_breadcrumbs_color":"","ocean_breadcrumbs_separator_color":"","ocean_breadcrumbs_links_color":"","ocean_breadcrumbs_links_hover_color":"","ocean_display_footer_widgets":"default","ocean_display_footer_bottom":"default","ocean_custom_footer_template":"","ocean_post_oembed":"","ocean_post_self_hosted_media":"","ocean_post_video_embed":"","ocean_link_format":"","ocean_link_format_target":"self","ocean_quote_format":"","ocean_quote_format_link":"post","ocean_gallery_link_images":"on","ocean_gallery_id":[],"footnotes":""},"categories":[286,169],"tags":[],"class_list":["post-7715","post","type-post","status-publish","format-standard","hentry","category-fractualites-ennews-fr","category-seminaires-armedia","entry"],"_links":{"self":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7715","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/comments?post=7715"}],"version-history":[{"count":1,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7715\/revisions"}],"predecessor-version":[{"id":7716,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7715\/revisions\/7716"}],"wp:attachment":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/media?parent=7715"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/categories?post=7715"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/tags?post=7715"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Mme C\u00e9cile MALLET<\/strong>, Professeure, Universit\u00e9 de Versailles Saint-Quentin-en-Yvelines, FRANCE – Rapporteur
M. Vincent AUGUSTO<\/strong>, Professeur, \u00c9cole des Mines de Saint-\u00c9tienne, FRANCE – Rapporteur
M. Eric VIBERT<\/strong>, Professeur des universit\u00e9s – praticien hospitalier, Universit\u00e9 Paris Saclay, FRANCE – Examinateur
M. J\u00e9rome BOUDY<\/strong>, Professeur, Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis, FRANCE – Examinateur<\/p>\n\n\n\n\u00ab Pr\u00e9diction personnalis\u00e9e du risque de mortalit\u00e9 sur liste d’attente en transplantation h\u00e9patique par apprentissage automatique explicable et transfert de connaissances \u00bb<\/h2>\n\n\n\n
pr\u00e9sent\u00e9 par Monsieur Abdelghani HALIMI<\/h2>\n\n\n\n
Abstract :<\/strong><\/p>\n\n\n\n