EXPERIENCE IN DIFFERENT FIELDS ALLOWS DATAMIND TEAM TO RAPIDLY IDENTIFY THE BEST TECHNOLOGIES AND METODOLOGIES FOR KNOWLEDGE EXTRACTION IN MEDICAL IMAGING.
Medical imaging is the technique and process used to create images of the human body (or parts and function thereof) for clinical purposes (medical procedures seeking to reveal, diagnose, or examine disease) or medical science (including the study of normal anatomy and physiology).
DataMind has worked in a few projects in this field, for example:
Multi-modality Organ Segmentation - Ultrasound (US) Tomography (CT)
Segmentation of targets and organs at risk in patient images used in radiotherapy is a very delicate task as errors in this phase propagate systematically through the whole treatment course. Moreover, it is becoming more and more time demanding since advanced treatment techniques, such as intensity modulated radiation therapy, were introduced. To improve workflow efficiency, semi-automated contouring techniques were implemented in treatment planning systems, based on computed tomography (CT) imaging datasets. But information from this imaging modality is not always sufficient to provide the necessary accuracy, even for manual contouring. This led to the need of using multiple imaging modalities in this process. Until now, no real automated multi-modality segmentation algorithm was available. In this work we introduce a three dimensional (3D) cross-modality automated image segmentation algorithm based on CT and ultrasound (US) images. The algorithm can be trained and optimized on the characteristics of specific patient samples, becoming increasingly accurate with the size of the sample. We cross-trained and cross-tested the algorithm on sixteen prostate patients. The conformity between the automatically segmented prostate contours and the contours manually outlined by an expert physician on the CT-US fusion was assessed using the Mean Distance to Conformity. In all of these cases the addition of US information reduced the MDC index. The fully automated cross-modality algorithm provided therefore contours that always matched the manually outlined references.
This project has been realized in collaboration with Tecnologie Avanzate Srl.
Lesion Detection - Tumor Delineation on PET
Radiation therapy is a local treatment, whose goal is to kill all clonogenic tumor cells and to reach a complete local tumor remission. Traditionally, treatment planning, monitoring and evaluation of response after radiotherapy are mainly based on computed tomography and magnetic resonance. However, concerning target volume delineation for radiotherapy, anatomical imaging has limits in the presentation of tumor extension, when tumor and normal tissues have similar density, similar magnetic properties or a similar contrast enhancement. Therefore, the new imaging methods of biological imaging such as 18F- fluorodeoxyglucose (FDG)-PET/CT imaging are increasingly used to define target volume in radiation treatment planning, although a standardized way of converting PET signals into target volumes is not yet available. To address these issue we developed an automated algorithm that delineates the target volumes determining dynamically the 3D region that minimizes a cost function extrapolated from the results of an important multicentric study in which the major Italian clinics were involved. The algorithm can handle volumes coming from all PET scanners and all kind of not-moving lesions.
This project has been realized in collaboration with Dott. Marco Brambilla, Secretary General of EFOMP, European Federation of Organisations for Medical Physics, Direttore SC di Fisica Sanitaria, Az. Ospedaliero Universitaria Maggiore della Carità di Novara and with Tecnologie Avanzate Srl.