Analytics for computation and visualization of liver resections

I nesten to tiår har datamaskiner hjulpet kirurger med å utføre mer presise operasjoner. Som i et videospill bruker leverspesialister 3D-modeller av leveren, generert fra ekte pasientdata, ikke bare for å se plasseringen og størrelsen på levertumorer, men også for å planlegge den kirurgiske strategien som passer best for pasienten (ved å interaktivt manipulere disse modellene). Takket være kunstig intelligens kan disse 3D-modellene nå lages raskere og mer automatisk. I ALive utvikler vi nye muligheter for å bruke 3D-modeller til å lage bedre kirurgiske planer: Hvordan kan vi lage 3D-interaktive verktøy som lar kirurger "designe" den beste kirurgiske planen i 3D? Kan datamaskinen lære hvordan en kirurg "designer" en plan og selv skape en? Mens 3D-virtuelle miljøer kan hjelpe mennesker med å forstå geometriske forhold mellom 3D-objekter (som organer og svulster), foretrekker vi i noen sammenhenger å ha informasjonen forenklet. Tenk på et kart, som er en forenkling av en 3D-verden, og som gjør det mulig for folk raskt å finne nyttig informasjon. På samme måte ønsker vi i ALive å lage forenklinger av de 3D-virtuelle reseksjonsplanene i 2D-diagrammer som kan hjelpe eksperter med å tolke en reseksjonsplan; 2D-diagrammer, i motsetning til 3D-modeller, kan brukes til å dokumentere reseksjoner i medisinske journaler og rapporter. For å nå disse målene lager vi i ALive-prosjektet mange 3D-modeller fra ekte pasienter, og sammen med kirurger designer vi de beste verktøyene for å lage reseksjonsplaner i 3D. Dermed kan vi, samtidig som vi gir nyttige 3D-verktøy til kirurgene, samle data om hvordan de planlegger reseksjoner. Denne informasjonen brukes videre til å trene algoritmer for kunstig intelligens til å lære prosessen og prøve å gjenskape den. Samtidig kan vi bruke reseksjonene til å studere den beste måten å forenkle reseksjonene i 2D-diagrammer.

Enhanced Liver Segmentation and Modeling: Developed AI algorithms, specifically a dual-encoder Y-shape network, that significantly improved volumetric liver and lesion segmentation, enabling automated 3D model generation critical for surgical planning. This advancement potentially improves surgical outcomes by allowing more precise and effective resections for liver cancer patients. Innovative Resection Planning Methods: Introduced novel computational methods for liver resection planning, including spline-based approximations inspired by laparoscopic demarcation. These methods improve the definition and planning of resections in complex scenarios involving multiple metastases. As a result, surgeons can plan complex resections more effectively, potentially leading to better patient outcomes. Advanced Vascular Modeling: Created parameterized vascular models integrating both portal and hepatic venous systems, surpassing conventional approaches that use only the portal system. This provides greater flexibility and accuracy in resection planning. This enhancement deepens clinical understanding of liver anatomy, influencing surgical decision-making and potentially improving patient care. Resectograms for Visualization: Developed "Resectograms," compact 2D visualizations extracted from 3D models, enhancing the visualization and understanding of surgical plans. This tool shows promise in optimizing resection plans and preventing malformed resections. By improving visualization, Resectograms aid surgeons in planning precise resections, which can lead to better surgical outcomes. Software Implementation in Slicer-Liver: Implemented all methods into Slicer-Liver, an open-source, user-friendly software distributed through 3D Slicer. This platform enhances optimization, visualization, and usability, facilitating worldwide collaboration and adoption. The accessibility of Slicer-Liver promotes global collaboration and training, accelerating advancements in liver surgery planning. Clinical Studies and Insights: Participated in multiple international clinical studies, providing valuable insights into liver surgery practices, tumor characteristics, and surgical techniques. This collaboration ensured the technological developments remained aligned with clinical needs. Aligning tools with clinical practice increases the likelihood of adoption and ultimately improves patient care. Advancements in Mixed-Reality Visualization: Contributed to integrating virtual surgical plans with mixed-reality devices like the Microsoft HoloLens, paving the way for future innovations in surgical visualization. This progress opens new avenues for enhancing surgical training and planning, potentially improving surgical precision.

Liver cancer is one of the most common types of cancer and its incidence is increasing. Surgical resection is the only curative treatment for some types of cancer. For nearly two decades, surgeons have been employing computer-assisted planning systems (CAPS); these systems show an increase of precision in surgical planning and an improved orientation and confidecne of the surgeon during operation. Despite these benefits, CAPS have found difficulties to Reach the clinical practice (the most noticeable is the problems to generate 3D patient-specific models from images). With the introduction of AI in medical imaging, these problems have been greatly reduced. This has created a new scenario where 3D patient-specific models are going to be systematically generated for its use in surgical planning and guidance. This new reality is perfect for the introduction of liver analytics and AI in surgical planning for the improvement of liver surgery practice. The main problems that can benefit from the introduction of AI and liver analytics are: (1) the difficulties for generation of resection plans in difficult cases (e.g., multiple metastases)—this proces is still manual—; (2) the standard division of the liver in segments (largely used for resectio planning) does not pose a wide consensus in the medical community, and therefore, there is the need to investigate new methods that can computationally generate different types of vascular territories; and (3) there are no formal methods to specify and communicate resection plans—clinicians are currently using subjective descriptions (written or oral), hand-drawings and pictures taken from the surgery, and therefore, there is a need for investigating visualization techniques able to capture the critical information contained in a resection plan in a formal way that can be interpreted by any clinical expert. This project will develop algorithms to solve these problems using analtyics, geometric modeling, visualization and AI.