Dynamic curriculum learning improves brain imaging classification and prediction

A dynamic curriculum learning framework for spatiotemporal encoding outperformed existing methods on six brain imaging tasks including Alzheimer's and tumor classification.

• DCL-SE consistently outperformed existing methods across six public neuroimaging datasets.
• The approach uses a data-driven spatiotemporal encoding without threshold-controlled stage-switching.
• Emphasizes the value of compact, task-specific architectures.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.