Anatomical and radiological evaluation of frontal lobe morphometry in healthy and dementia people and machine learning-based prediction of dementia

Creative Commons License

Özandaç S., Tunç M., Öksüzler M., ÇOBAN Ö., Özel S. A., Karakaş P.

Cukurova Medical Journal, vol.48, no.2, pp.541-558, 2023 (ESCI) identifier

  • Publication Type: Article / Article
  • Volume: 48 Issue: 2
  • Publication Date: 2023
  • Doi Number: 10.17826/cumj.1275723
  • Journal Name: Cukurova Medical Journal
  • Journal Indexes: Emerging Sources Citation Index (ESCI), Academic Search Premier, Directory of Open Access Journals, TR DİZİN (ULAKBİM)
  • Page Numbers: pp.541-558
  • Çukurova University Affiliated: Yes


Purpose: This paper aimed to determine the morphometry of the frontal lobe and central brain region using magnetic resonance imaging in patients having dementia and healthy subjects. Materials and Methods: 243 subjects (121 subjects having dementia; 122 subjects healthy group) aged 60-90 years over for 2 years between January 2018 and 2020 were included in this study. Also, the supervised Machine learning based (ML based) detection of dementia has been studied on this obtained real world data. Results: The gender-related changes of frontal region measurements in dementia and healthy subjects were analyzed and, there were differences of measurements’ mean values in gender. In healthy subjects, significance differences were found in all measurements (except the distance from anterior commissure to posterior commissure and outermost of corpus callosum genu to innermost of corpus callosum genu). The means of the measurements were found higher in males than in females. Conclusions: We believe that the knowledge of our study will provide valuable reference data for our population and will help for a surgeon in planning an operation by considering measurements related to the frontal lobe. In addition, ML based supervised methods that were trained on the collected data for detection of dementia showed that it is required to provide as many attributes and instances as possible to train an accurate estimator. However, if this is not possible, by creating new features based on the hidden patterns between attributes and instances we could increase the success of the estimators up to 96.3% f-score value.