Differentiating Idiopathic Normal Pressure Hydrocephalus Using Multi-Head Attention CNN with Diversify-Loss in Brain MRI

Idiopathic Normal Pressure Hydrocephalus (iNPH) is characterized by excessive accumulation of cerebrospinal fluid in the cerebral ventricles, leading to symptoms such as gait disturbances, cognitive impairments, and urinary incontinence. Shunt surgery can significantly improve these symptoms. Conversely, Progressive Supranuclear Palsy (PSP) presents similar symptoms but lacks effective treatment options, making accurate differentiation crucial. Traditional methods, which rely on visual assessment of brain MRI, can be challenging without a specialist and are prone to intra- and inter-observer variability. This study proposes an automated approach using a Multi-head Attention Convolutional Neural Network (MA-CNN) with Diversify-Loss to differentiate iNPH from PSP. The proposed method was validated on 39 iNPH and 39 PSP subjects and achieved an accuracy of 0.962 and an Area Under the Curve (AUC) of 0.978, significantly outperforming both the conventional CNN (accuracy of 0.822 and AUC of 0.936) and the method without Diversify-Loss (accuracy of 0.911 and AUC of 0.966). These results indicate that automated differentiation is feasible and that attention maps coupled with Diversify-Loss significantly enhance diagnostic accuracy.