At CESP, Vincent-Béni Sèna Zossou has developed an AI tool for liver cancer detection.

Vincent-Béni Sèna Zossou, chercheur·e au sein de CESP (Université de Versailles Saint-Quentin-en-Yvelines).

Thèse soutenue en 2024 à l'école doctorale École doctorale Santé Publique (Le Kremlin-Bicêtre, Val-de-Marne ; 2015-...).

Référencée dans le réseau ABES/STAR, cette thèse répond aux critères de rigueur de l'enseignement supérieur français.

Early detection of hepatocellular carcinoma (HCC) is a major public health issue, particularly in West Africa, where the incidence rates of this disease are alarming. Indeed, HCC is often diagnosed at an advanced stage, which significantly limits treatment options and increases mortality rates. Traditional diagnostic methods, which rely on visual analysis of computed tomography (CT) images, are often subject to human errors and biases, which can lead to misdiagnoses. In this context, Vincent-Béni Sèna Zossou, within the Center for Studies and Services in Public Health (CESP), proposes an innovative solution by integrating artificial intelligence (AI) to automate this crucial process. This research addresses an urgent need for improved diagnostics in Africa, where medical resources are often limited and access to advanced technologies remains a challenge.

The use of machine learning (ML) allows for the processing of a large volume of CT images, thereby reducing the time required to establish a diagnosis. Indeed, AI algorithms can analyze thousands of images in a matter of minutes, providing a speed of execution that is essential in contexts where every minute counts. The preliminary results of this research show Dice scores ranging from 69% to 97.1%, indicating promising performance of the developed segmentation models. However, the question remains: how can we ensure that these tools are adapted to the specificities of African populations, which may present clinical and radiological characteristics different from those observed in other regions of the world?

The methodology adopted by Zossou is based on the collection of CT images and their annotation by experienced radiologists. This step is crucial for creating a reference dataset, essential for training AI models. By integrating regularization techniques, it is possible to improve the robustness of the models against data variations, whether due to differences in imaging equipment or variations in clinical practices. This raises a reflection on the importance of continuous training for healthcare professionals in Africa to maximize the impact of these technologies. Indeed, it is imperative that radiologists and other healthcare professionals are trained not only in the use of these tools but also in their interpretation, to ensure a smooth integration into daily clinical practices.

The strategic recommendations arising from this research go beyond mere technical improvement. They highlight the need to develop diagnostic solutions that meet specific clinical needs in Africa. This involves close collaboration between researchers, clinicians, and policymakers to ensure that the tools developed are not only effective but also accessible and usable in the local context. For example, initiatives could be established to create regional training centers where healthcare professionals could acquire skills in medical imaging and AI.

Moreover, it is essential to take into account existing healthcare infrastructures. In many regions of West Africa, access to high-quality CT equipment is limited. Therefore, the solutions developed must be adapted to these realities, considering technical and financial constraints. This could include the development of AI models capable of functioning with images of varying quality, or the establishment of partnerships with international organizations to improve access to technology.

In summary, this thesis represents a significant advancement in the field of medical imaging in West Africa, with considerable potential impact on public health. By integrating artificial intelligence into the detection process of hepatocellular carcinoma and liver metastases, we have the opportunity to transform the landscape of medical diagnosis in the region. This could not only improve early detection rates but also contribute to better management of health resources, cost reduction, and, most importantly, improved patient outcomes. It is therefore crucial to continue this research while ensuring that the solutions developed are truly adapted to African realities.

Sources et accès

Vincent-Béni Sèna Zossou. Détection du carcinome hépatocellulaire et des métastases hépatiques basée sur les images tomodensitométriques et l'apprentissage automatique. Cancer. Université Paris-Saclay; Université d'Abomey-Calavi UAC (Bénin), 2024. Français. ⟨NNT : 2024UPASR034⟩. ⟨tel-05016575⟩

• Fiche HAL complète
• Texte intégral (PDF)
• Publication éditeur — DOI
• Theses.fr — NNT 2024UPASR034