OBJECTIVE: The process of diagnosing many neurodegenerative diseases, such as Parkinson's and progressive supranuclear palsy, involves the study of brain magnetic resonance imaging (MRI) scans in order to identify and locate morphological markers that can highlight the health status of the subject. A fundamental step in the pre-processing and analysis of MRI scans is the identification of the mid-sagittal plane, which corresponds to the mid-brain and allows a coordinate reference system for the whole MRI scan set. APPROACH: To improve the identification of the mid-sagittal plane we have developed an algorithm in Matlab® based on the k-means clustering function. The results have been compared with the evaluation of four experts who manually identified the mid-sagittal plane and whose performances have been combined with a cognitive decisional algorithm in order to define a gold standard. MAIN RESULTS: The comparison provided a mean percentage error of 1.84%. To further refine the automatic procedure we trained a machine learning system using the results from the proposed algorithm and the gold standard. We tested this machine learning system and obtained results comparable to medical raters with a mean absolute error of 1.86 slices. SIGNIFICANCE: The system is promising and could be directly incorporated into broader diagnostic support systems.
OBJECTIVE: The process of diagnosing many neurodegenerative diseases, such as Parkinson's and progressive supranuclear palsy, involves the study of brain magnetic resonance imaging (MRI) scans in order to identify and locate morphological markers that can highlight the health status of the subject. A fundamental step in the pre-processing and analysis of MRI scans is the identification of the mid-sagittal plane, which corresponds to the mid-brain and allows a coordinate reference system for the whole MRI scan set. APPROACH: To improve the identification of the mid-sagittal plane we have developed an algorithm in Matlab® based on the k-means clustering function. The results have been compared with the evaluation of four experts who manually identified the mid-sagittal plane and whose performances have been combined with a cognitive decisional algorithm in order to define a gold standard. MAIN RESULTS: The comparison provided a mean percentage error of 1.84%. To further refine the automatic procedure we trained a machine learning system using the results from the proposed algorithm and the gold standard. We tested this machine learning system and obtained results comparable to medical raters with a mean absolute error of 1.86 slices. SIGNIFICANCE: The system is promising and could be directly incorporated into broader diagnostic support systems.
Authors: Rossella Rizzo; Silvin P Knight; James R C Davis; Louise Newman; Eoin Duggan; Rose Anne Kenny; Roman Romero-Ortuno Journal: Geriatrics (Basel) Date: 2022-04-22