Machine Learning

Classification of Alzheimer’s Disease Stages using Radiology Imaging and Longitudinal Clinical Data – Part 10


The project has fully answered research questions after implementing different machine learning techniques and evaluating the performance of the models. The developed model and user-friendly web application contribute to supporting an end-to-end pipeline to predict the stages of Alzheimer’s disease. This is a continuation from here.

Further, the practical application, use of libraries, software contribute to the innovation in the research of the disease. The project consolidates the findings from the literature review (Green et al., 2011), (Lee et al., 2016), (Goyal et al., 2018), (Venkatraghavan et al., 2019), (Wang et al., 2019) that radiology images and longitudinal clinical data can help to effectively and efficiently predict disease progression. The project successfully implements the most recent techniques such as building a machine learning classifier, explaining the output of the models, publishing the code to GitHub and deploying a functional web-based application. The model also handles missing data to improve the performance of the model. It supports the various features from ADNI data set to provide a data-driven approach to model the disease progression.

The project finds cognitive tests like CDRSB, MMSE, MRI of the whole brain and other factors such as age to be important features to predict the stages. The importance of cognitive tests is in accordance with other papers such as (Goyal et al., 2018) and (Mehdipour Ghazi et al., 2019) but different than the findings of a paper (Lee et al., 2016). The practical experience has resulted in the researcher gaining new skills in Python programming, data manipulation, visualization and machine learning techniques.

However, it needs to be stressed that the project aims to act as a supplementary tool to be used in conjunction with the skills and knowledge of the clinical staff. The models are simplified and find the major patterns in the data. The results from the model need to be verified by the physician. Supervised classification approach requires many labelled data thus making it dependent on the ground truth provided by the diagnosis from the clinicians. Hence, it does not consider any misdiagnosis. Further, a single sequence of features is used to predict the stage of the progression but sometimes the rate of progression is unpredictable. The project uses historical longitudinal data which have been pre-processed to classify the different stages. In future, unsupervised classification algorithms e.g., k-means clustering can be applied to improve reliability.

Moreover, it uses the data set from a challenge. The model can also be utilized on a different data set to ensure the ability to transfer the techniques. Furthermore, the model classifies only three stages namely normal, MCI and dementia. The granularity of the disease can be further studied and modelled as future work. Despite its limitations, the project showcases a framework that has clear importance as exhibited by the impressive performance of the model. The proposed ensemble of classifiers includes feature selection component and results in good classification after testing on unseen data.

Conclusion and Future Work

The main aim of the project is to develop a model and web-based application to predict the different stages of Alzheimer’s disease. The gaps are identified through the literature review and then necessary steps are taken to address these gaps. Different machine learning techniques are used to build a multiclass classifier. An ensemble of classifiers is found to be the best model. A web-based application is also deployed to further the research and enable the user to predict the occurrence of the disease. It is important to mention that all the objectives mentioned in section 1.3 have been achieved.

Future Work

The proposed approach can be improved in the future in the following ways:

  • The resulting predictions just indicate the three stages of Alzheimer’s disease. This can be extended to more granular stage
  • The predictions can be more probabilistic in nature to include the likelihood of diagnosis and present the next course of action for treatment.
  • The results of the model can be logged and monitored to find the performance of the machine learning model over time.
  • The web-based application is simple and provides a proof of concept. The features and model used to create can be improved further along with enhanced user design and experience.


Leave a Reply