Abstract
Introduction:
Head and neck cancers (HNC) are the second most common cancers among Indian population and most common cancer among Indian men. The autosegmentation algorithm of Siemens Healthineers version AI-Rad Companion VA 30 (AASH) for head and neck nodal contouring was recently released. This is the first study to evaluate this AI algorithm in our network of cancer centers.
Materials and Methods:
A total of 108 patients of various subsites of HNC treated during the period of February 2023–May 2023 were analyzed. Six physicians from different locations of the network hospitals participated in the study. The automated nodal contours from level IA to VIIB generated by AASH were evaluated using 4 points scale—a score of 4 is clinically usable with no edits, a score of 3 requires minor edits, a score of 2 requires major edits, and a score of 1 requires complete re-contouring of the region. Physicians manually recorded time to contour manually or review AI contours. IBM SPSS Statistics, version 27, was used to perform all statistical analyses.
Results:
Of 108 patients with median age of 55 years, 78.7% were male, 54.5% were node positive, and 53.7% had definitive RT. The mean score of all nodal group was 3.56 ± 0.40 with mean scores for level IA, IB, II, III, IVA, IVB, V, VIA, VIB, VIIA, and VIIB being 3.31, 3.53, 3.63, 3.7, 3.74, 3.75, 3.68, 3.7,3.63, 2.55, and 3.78, respectively. Level VIIA had lowest score with scores being 3 and 4 only 60% of times compared to >80% of most other levels. On multivariate analysis, the only factor which predicted variation in score was physician. The mean time to contour manually versus AI was 18.6 versus 5.05 min (p = 0.01).
Conclusion:
Overall, the AI-segmented autocontouring performed well with significant time-saving and was clinically usable with no or minor edits majority of times for all levels except VIIA. There is some variation in scoring based on individual physician preference.
Get full access to this article
View all access options for this article.