Pramita, Levina Lintang Pramita (2025) Prediction of alzheimer\'s disease based on mri data using machine learning cnn with the detrac method / Levina Lintang Pramita</p>. Diploma thesis, Universitas Negeri Malang.
Full text not available from this repository.Abstract
p Alzheimer s disease (AD) is a neurodegenerative disorder that causes cognitive impairments such as memory loss (Knopman dkk. 2021). In 2006 the number of AD patients reached 26.6 million and is projected to quadruple by 2050 (Brookmeyer dkk. 2007). Early detection is crucial to slowing the progression of AD yet several challenges hinder its diagnosis including limited consultation time for doctors difficulties in achieving accurate diagnoses and the misconception that AD symptoms are a normal part of aging (Porsteinsson dkk. 2021). Additionally identifying the stages of AD is essential in determining suitable treatment for patients (Burke amp Goldfarb 2022). Therefore an Artificial Intelligence (AI)-based prediction system is needed to facilitate more efficient and accurate early diagnosis and stage progression mapping of AD. This study aims to implement the DeTraC (Decompose Transfer Learning Compose) method for predicting AD based on MRI images and evaluate its performance. The method combines Transfer Learning (TL) and Class Decomposition (CD) to improve model accuracy despite limited datasets. The research methodology consists of six main stages (1) literature review on AD classification (2) data collection of Alzheimer s MRI images from Kaggle (3) data preprocessing including cropping to remove irrelevant outer brain regions and oversampling to balance the number of samples across classes (4) designing solution method (5) developing solution method and (6) model evaluation using accuracy precision recall and F1-score metrics. Experiments were conducted using four different scenarios (no subclasses two subclasses three subclasses and four subclasses). Experimental results indicate that class decomposition has impacts on model performance. The baseline model (without subclass decomposition) stabilized at epoch 20 with an accuracy of 98%. The two-subclass model achieved the highest accuracy of 99% at epoch 25 demonstrating that class decomposition can help the model learn more specific patterns. In the three-subclass scenario the model required 40 epochs but its accuracy dropped to 97%. Meanwhile in the four-subclass scenario even though precision recall and F1-score remained at 99% the accuracy decreased back to 98% requiring 50 epochs. These findings indicate that increasing the number of subclasses does not necessarily enhance performance instead it can increase model complexity without providing significant benefits. Additionally oversampling proved effective in addressing class imbalance particularly in improving the model rsquo s ability to predict data in the Moderate Demented and Mild Demented classes which originally had the fewest samples. Based on the research findings it can be concluded that the implementation of two subclasses is the optimal configuration in this study as it provides a balance between model stability and high performance. The two-subclass model reached stability at epoch 25 achieving an accuracy precision recall and F1-score of 99%. Selecting the optimal number of subclasses is a crucial factor in avoiding increased complexity and performance degradation. This model was further evaluated using 10-fold cross-validation yielding an average precision of 98.3% recall of 97.8% F1-score of 98.2% and accuracy of 97.7%. Compared to previous studies these results demonstrate that DeTraC is a more effective approach in Alzheimer rsquo s disease classification. For future research it is recommended to explore different pretrained models such as ResNet EfficientNet Inception or other architectures to compare their effectiveness with DeTraC. Additionally experiments with different datasets are necessary to assess whether the DeTraC method can be applied to other medical cases. Further hyperparameter optimization can be conducted to improve model performance. /p
| Item Type: | Thesis (Diploma) |
|---|---|
| Divisions: | Fakultas Teknik (FT) > Departemen Teknik Elektro (TE) > S1 Teknik Informatika |
| Depositing User: | library UM |
| Date Deposited: | 22 Apr 2025 04:29 |
| Last Modified: | 09 Sep 2025 03:00 |
| URI: | http://repository.um.ac.id/id/eprint/426194 |
Actions (login required)
 |
View Item |