Combining convolution neural networks with long‐short term memory layers to predict Parkinson's disease progression

Abstract Parkinson's disease (PD) is a neurological condition that occurs in nearly 1% of the world's population. The disease is manifested by a sharp drop in dopamine production, resulting from the death of the related producing cells in an area of the midbrain called the substantia nigra. Early diagnosis and accurate staging of the disease are essential to apply the appropriate therapeutic approaches to slow cognitive and motor decline. At present, there is not a singular blood test or biomarker accessible for diagnosing PD or monitoring the progression of its symptoms. Clinical professionals identify the disease by assessing the symptoms, which, however, may vary from case to case, as well as their progression speed. Magnetic resonance imaging (MRIs) have been used for the past three decades to diagnose and distinguish between PD and other neurological conditions. However, to the best of our knowledge, no neural network models have been designed to identify the disease stage. This paper aims to fill this gap. Using the “Parkinson's Progression Markers Initiative” dataset, which reports the patient's MRI and an indication of the disease stage, we developed a model to identify the level of progression. The images and the associated scores were used for training and assessing different deep learning models. Our analysis distinguished four distinct disease progression levels based on a standard scale (Hoehn and Yah scale). The final architecture consists of the cascading of a 3D‐CNN network, adopted to reduce and extract the spatial characteristics of the MRI for efficient training of the successive LSTM layers, aiming at modeling the temporal dependencies among the data. Before training the model, the patient's MRI is preprocessed to correct acquisition errors by applying image registration techniques, to extract irrelevant content from the image, such as nonbrain tissue (e.g., skull, neck, fat). We also adopted template‐based data augmentation techniques to obtain a balanced dataset about progression classes. Our results show that the proposed 3D‐CNN + LSTM model achieves state‐of‐the‐art results by classifying the elements with 91.90 as macro averaged OVR AUC on four classes.