• registration
• medical imaging
• machine vision
• computer assisted surgery
• segmentation

This thesis proposes methods and procedures to proficiently introduce patient 3D models in the daily clinical practice for diagnosis and treatment of abdominal diseases. The objective of the work consists in providing and visualizing quantitative geometrical and topological information on the anatomy of interest, and to develop systems that allow to improve radiology and surgery.
The 3D visualization drastically simplifies the interpretation process of medical images and provides benefits both in diagnosing and in surgical planning phases. Further advantages can be introduced registering virtual pre-operative information (3D models) with real intra-operative information (patient and surgical instruments). The surgeon can use mixed-reality systems that allow him/her to see covered structures before reaching them, surgical navigators for see the scene (anatomy and instruments) from different point of view and smart mechatronics devices, which, knowing the anatomy, assist him/her in an active way. All these aspects are useful in terms of safety, efficiency and financial resources for the physicians, for the patient and for the sanitary system too.
The entire process, from volumetric radiological images acquisition up to the use of 3D anatomical models inside the surgical room, has been studied and specific applications have been developed.
A segmentation procedure has been designed taking into account acquisition protocols commonly used in radiological departments, and a software tool, that allows to obtain efficient 3D models, have been implemented and tested.
The alignment problem has been investigated examining the various sources of errors during the image acquisition, in the radiological department, and during to the execution of the intervention. A rigid body registration procedure compatible with the surgical environment has been defined and implemented. The procedure has been integrated in a surgical navigation system and is useful as starting initial registration for more accurate alignment methods based on deformable approaches.
Monoscopic and stereoscopic 3D localization machine vision routines, using the laparoscopic and/or generic cameras images, have been implemented to obtain intra-operative information that can be used to model abdominal deformations. Further, the use of this information for fusion and registration purposes allows to enhance the potentialities of computer assisted surgery. In particular a precise alignment between virtual and real anatomies for mixed-reality purposes, and the development of tracker-free navigation systems, has been obtained elaborating video images and providing an analytical adaptation of the virtual camera to the real camera.
Clinical tests, demonstrating the usability of the proposed solutions, are reported. Test results and appreciation of radiologists and surgeons, to the proposed prototypes, encourage their integration in the daily clinical practice and future developments.