Optimierung und klinische Evaluation einer intelligenten Atlas-basierten semiautomatischen Zielvolumendefinition am Beispiel von Kopf-Hals-Tumoren

The aim of this work was to optimize in constant cooperation with the department of Informatics of the "Rhein-Main-Hochschule" a method for (semi-)automatic target volume definition using the example of head and neck cancers, in special the CTV 3 of laryngeal carcinoma, and to evaluate the use in clinical practice. The developed method is based on building an atlas consisting of a target volume and a landmark model of n-CT data sets. This approximate atlas volume can be transferred by an affine transformation on an individual patient data set, in which defined landmarks are marked. The landmarks describe characteristic reproducible anatomical structures, that represent the individual anatomy of the patient. First, 10 CT data sets of patients with head and neck cancer were selected. In each of these data sets one pre-defined reference target volume and the socalled landmarks, which were defined on different levels of the target volume, were drawn. Afterwards a target volume atlas was generated. The Similarity Index, that quantifies the over-lapping of two volumes as a percentage, is used to compare the different volumes. The originally defined landmarks have been improved by a systematic reduction method and clinical assessment. So the Similarty Index of 73% of the 16 landmarks could be increased to 77 % for 13 optimized landmarks. To evaluate the clinical applicability of (semi-)automatic target volume definition 5 physicians, who are experienced in contouring target volumes in head and neck region, manually drew the target volume for locally advanced laryngeal carcinoma in two different CT data sets. They also marked the optimized landmarks in these two CT data sets. A target volume was then automatically generated by using the atlas. This automatically generated volume was then evaluated and adjusted by the physician who marked the landmarks. So a semiautomatic created target volume resulted. In comparison to the reference volume the average similarity index of the manual contouring was 76%, the one of the automatic target volume delineationwas 74% and the one of the semi-automatic method was 79%. Manual contouring required an average duration of 28.7 minutes, whereas the automatic atlas based target volume definition was finished in 2.4 minutes and the semiautomatic method in 13.8 minutes. So a time saving of about 52 % for the semiautomatic delineation method and about 91% for automatic contouring in comparison with the manual delineation could be achieved. The comparisons of the delineated target volumes within the 5 physicians showed Similarity Indices for the manual volume definition from 64% to 80 % and for the semi-automatic contouring from 73 to 83%. This shows that the interobserver variability is reduced by using the semi-automatic atlas-based method. In summary, it was shown that atlas-based semi-automatic target volume definition in head and neck cancer is clinically applicable and combines the advantages of time saving and reduced interobserver variability.