Reconstruction of Imagined Melody with Relative Pitch Decoding in Electrocorticography

Music imagery involves internally generated auditory experiences, yet decoding imagined melodic content from neural activity remains challenging. Here, we examined whether imagined melodies can be decoded from human electrocorticography (ECoG) recordings using a relative pitch framework. Ten epilepsy patients performed a music imagery task involving familiar melodies presented in multiple tonalities (3 males and 7 females; mean age, 27.3 ± 4.69 years). Neural features from imagery-responsive cortical sites were used to train intrasubject models to decode relative pitch classes at the single-note level. Although single-note decoding accuracy was modest, it reliably exceeded chance. Sequential integration of decoded pitch classes enabled reconstruction of melodic contours that preserved the relative structure of imagined melodies. These results provide a proof-of-concept that imagined melodic structure can be accessed from intracranial neural recordings when framed in terms of relative pitch, underscoring the importance of sequence-level approaches for decoding internally generated auditory content.