This study aimed to explore that if the segmentation of different volume
of interests (VOIs) may influence the diagnostic performance of radiomic model.
We included 78 patients with pathologically confirmed uterine sarcomas or
atypical leiomyomas. 3 different VOIs were manually drawn on images of ADC maps.
Radiomic models were built based on three feature set. Features extracted from VOI
covered the whole uterus had the best diagnostic performance than VOI covered
the lesion or lesion and some surrounded tissue. It suggested VOI covered the whole uterus added relevant information for distinguishing
uterine sarcoma from atypical leiomyoma.