Artificial intelligence and, particularly, machine learning have shown significant impact in diverse fields. Among them, medicine stands as one with tremendous potential along with equally critical challenges, as the increasing number of publications on image-based diagnosis — radiographs, histology and optic fundi, among others — shows.
Radiology is the field of medicine with the greatest number of clinical solutions that use AI. These solutions focus on detecting findings, classifying lesions, organ or lesion segmentation and image processing, among others. Segmentation is very common in medical imaging, with liver segmentation being one of the most explored because of its applications in volumetry. Liver segmentation for volumetry is indicated in patients undergoing major liver surgery, defined as the resection of four or more liver segments. Most hepatectomies are performed in patients with liver neoplasms, such as hepatocarcinoma, cholangiocarcinoma and liver metastases. Liver volumetry estimates the total liver volume and the future liver remnant: the amount of liver parenchyma left after resection. The future liver remnant volume is directly correlated with post-hepatectomy liver function, and measuring it precisely is crucial to prevent post-hepatectomy liver failure.
Current models focus on liver and tumour segmentation, an important task but not enough for surgical planning. An automatic liver and tumour segmentation method can greatly relieve physicians of the heavy workload of examining CT images; surgery, however, requires accurately estimating the remnant liver volume after resection, for instance in patients with liver metastases.
Estimating the future liver remnant is challenging because the type of surgery depends on each patient's clinical setting, the centre's experience, and the number and location of liver lesions, among others. Future liver remnant segmentation therefore depends on the patient's clinical context.
The goal of this project is to design, implement and evaluate fine-grained liver segmentation guided by context, so that the remnant liver volume can be estimated precisely. Specific objectives:
- Evaluate state-of-the-art liver segmentation models.
- Design and evaluate fine-grained liver segmentation models.
- Estimate the remnant liver volume using the fine-grained segmentation model.
- Integrate contextual information through prompts for liver segmentation.
- Evaluate and adjust the proposed models.

Researchers
- Héctor Sebastián Henríquez Leighton (Clínica Santa María, UANDES, Chile)
- Jose Manuel Saavedra Rondo (UANDES, Chile)
- Violeta Chang (USACH, Chile)
- Aline Xavier (USACH, Chile)
- Caroline Petitjean (Université de Rouen Normandie, France)
- Fannia Pacheco (Université de Rouen Normandie, France)
- Enzo Ferrante (CONICET, Argentina)
- María Vakalopoulou (CentraleSupélec, Paris, France)
Activities and events

Academic meeting with our French team at CentraleSupélec, Paris, focused on medical imaging research and international collaboration.


Meeting with colleagues from Université de Rouen Normandie to discuss one-shot image spotting and explainability techniques for medical imaging.


Visit to Universidad de Buenos Aires with Enzo Ferrante’s team, sharing activities and research topics in a Latin American collaborative environment.


Participation in the IV Workshop on Computer Vision, with keynote talks and strong support from local organizers and collaborators.