NAS-optimized topology-preserving transfer learning for differentiating cortical folding patterns

•This is the first attempt to utilize the topology-preserving transfer to differentiate cortical folds (sulci, 2HGs and 3HGs gyri).•Extensive experiments obtain consistent high performance in the two ASD datasets.•NAS-optimized network architecture obtains a higher accuracy over the baseline, extending the application of NAS.•Classification results demonstrate that the differences across individuals increase with age.