“It’s a population that’s just silently suffering”. These words by doctors and physicians for people suffering from dementia have inspired me to aid the process of detecting Alzheimer's disease in patients by using data mining and machine learning techniques. Traditionally, Alzheimer's disease was only diagnosed with complete certainty after the onset of the severe condition. Today, images of the brain are used chiefly to pinpoint visible abnormalities related to conditions other than Alzheimer's disease — such as strokes, trauma or tumours — that may cause cognitive change. These cognitive changes can be utilized to predict the stage of Alzheimer's disease on the CDR scale. No blood test, brain scan, or physical exam can definitively diagnose Alzheimer’s disease. And because so many conditions can produce symptoms resembling those of early Alzheimer’s, reaching the correct diagnosis is complicated. According to our analysis, these 4 features can help in predicting the CDR value for early detection from Brain MRI
What it does
The Xgboost trained model utilises a set of input parameters from the user which includes the Brain MRI scan information and the personal background information to predict the CDR value for early detection from Brain MRI.
How we built it
It is built through a processed stage involving-
- Data preprocessing
- Training Linear Model and analysing the performance, p-values, fitting plots, residuals, R-square values, etc.
- Training Xgboost Model
- Training Neural Network Model
- Used Plumber API to connect model from R studio to the Website
- Built the front end and configured the API
Challenges we ran into
In the analysis phase, the P values from the linear model indicated only nWBV and MMSE as the significant contributing factors. Hence while training the XG boost model is utilised only these two parameters. However, I soon realised that a person's age and other background related to environment and lifestyle are also important criteria is that all suspected by physicians and cannot be left behind in the model training. I retrained the model and extracted important features from XGboost trained model and was satisfied to see that this time it included a person's age and socio-economic background as important factors as well.
Accomplishments that we're proud of
It is an accomplishment to understand the data thoroughly and be able to tell what are the main features that exactly contribute to words the onset of Alzheimer's disease. Also, the Xgboost and Neural Network models have low MSE and the returning prediction values are very close to real values and hence satisfying😌. I stuck a lot of times on the extra boost model because it was not giving good results. I did hyperparameter tuning and also advanced to the neural network model to be able to improve my results for the end-user.
What we learned
I learnt a lot about Data mining and understood when to utilise a particular machine learning model by understanding the dataset. I also understood how to deploy the model from Rstudio to the website using Plumber API. Moreover, I also understood the difficulties that Alzheimer's patients face and I would like to continue in this field to help more and more people identify their condition timely and with more surety.
What's next for Saving Memories in Hearts
The next step for Saving Memories in Hearts is to provide a platform where people can upload the MRI images and from there the model is able to extract minute and important features and be able to predict the CDR score.
Log in or sign up for Devpost to join the conversation.