In this study, we investigated the
feasibility of differentiating AC from SCC using radiomics features extracted
from multiple MR series (T2TRA, T2SAG, ADC, CETRA and CESAG). The results
indicated that radiomics features identified by careful feature selection and
machine learning can have good performance for distinguishing AC from SCC. In
particular, T2SAG sequences had the best ability, followed by ADC and T2TRA
sequences, as demonstrated by both unsupervised clustering and supervised classification.
In general, we conclude that ACs have greater textural heterogeneity than SCCs, which was revealed