library_name: monai
tags:
- classification
- medical
- neuroscience
- alzheimer
license: mit
metrics:
- accuracy
pipeline_tag: image-classification
datasets:
- ADNI
Model Description
A machine learning model for waste classification
- Developed by: rootstrap
- Model type: image-classifier
- License: mit
Waste Classifier Model
The aim is to build a model for MRI classification that identifies among the different classes:
- Alzheimer's
- Mild Cognitive Impairment
- Control
This machine learning model will help medical staff get a second opinion on whether a pacient MRI indicates the presence of Alzheimer's desease
The model was built using Monai MONAI is a freely available, community-supported, PyTorch-based framework for deep learning in healthcare imaging.
It has two main design goals:
To be approachable and rapidly productive To be also configurable.
Model Sources
- Repository: https://github.com/rootstrap/MRI-classifier
Uses
At Rootstrap we classify waste. We found that people were struggled to classify correctly, and then we end up not recycling most of the generated waste at the office, since if there were items in the wrong basket, all the basket should not be classified. Because of this reason, we created an app to help people at our company to classify waste.
Direct Use
model = nets.DenseNet121(spatial_dims=3, in_channels=1, out_channels=3)
checkpoint = torch.load("86_acc_model.pth")
model.load_state_dict(checkpoint)
model.predict()
Bias, Risks, and Limitations
This model has an 86% of accuracy. Although it is an outstanding result, his means that 13% of times the prediction is wrong. This is why its important to understand that this tool is not a real medical opinion and can not be used as a final diagnosis by any means. This project does not aim to replace medical staff in diagnosing Alzheimer's desease, instead it is a tool to help them to get a quick and accurate second opinion.
Training Details
Training Data
The MRI data was gathered from the Alzheimer’s Desease Neuroimaging Initiative (ADNI) Database. Afterwards, the data was splitted into one folder for each class.
Split into train/test The data was splitted using the 60% for training, 20% for validation and 20% for testing
The model has been trained to classify waste into 3 classes. The dataset used for the training consisted of 2527 images:
- 501 glass
- 594 paper
- 403 cardboard
- 482 plastic
- 410 metal
- 137 trash
Training Procedure
You can find the code for training at train.ipynb Training the model using DenseNet121 adapted for 3D images.
Evaluation and Results
After 100 epochs, the model reached an accuracy of 85.76%