The invention discloses a
meningioma Ki67 classification method based on multi-
modal medical data, which is applied to the field of
image processing, and aims to solve the problem of overlarge calculation overhead during multi-
modal fusion in the prior art based on image data,
radiology data and
radiomics data of multi-
modal 3D MRI (three-dimensional
magnetic resonance imaging) during
meningioma Ki67 prediction. According to the method, iterative modal
information fusion is carried out through the structured
potential space and the cross attention mechanism, and the calculation complexity is remarkably reduced. According to the method, through dynamic
weight distribution and
iterative refinement, complementarity information between
modes is effectively captured, redundant
noise is suppressed, and the problem of feature
dilution caused by mode length difference or simple superposition in a traditional splicing method is avoided; the compact representation of the
potential space forces different modals to perform efficient interaction on the shared dimension, not only retains the key mode of cross-modal association, but also adaptively filters irrelevant contents through an attention mechanism, thereby realizing better balance on
information density and
semantic consistency, and improving the accuracy of the
information density and
semantic consistency. And the method is especially suitable for
processing real scenes with unbalanced
modes or
partial loss.