Connected Speech Classification

Classification of Alzheimer's disease and Amyloid Status from Transcribed Connected Speech.

Installation

To install the code, do the following:

pip install git+https://github.com/lcn-kul/connected-speech-classification.git

Requirements can be installed from the requirements.txt file doing something like this:

conda create --name connected-speech-classification python=3.11
pip install -r requirements.txt

Reproduction of Results

The analysis roughly consists of three steps: Data preparation, classification, formatting of the results.

Data Preparation

python ./connected_speech_classification/data/prepare_disease_status_datasets.py prepare-ad-hc-amyloid-pos-neg-datasets

Running the Models

1. Independent classification

1.1 Classification on AD vs half of amyloid negative HC

python ./connected_speech_classification/models/disease_status_classifier classify-disease-label

1.2 Classification of amyloid positive vs other half of amyloid negative HC

python ./connected_speech_classification/models/disease_status_classifier classify-disease-label --classify_amyloid

1.3 Classification of amyloid negative group 1 versus amyloid negative group 2

python ./connected_speech_classification/models/disease_status_classifier classify-disease-label --classify_baseline

2. Sequential classification Make sure to get the saved model "your-saved-model".

python ./connected_speech_classification/models/disease_status_classifier.py classify-disease-label \
	--classification_model_base "your-saved-model" \
	--classify_amyloid

3. Joint multi-task classification

In this case, you need to provide two dataset directories.

python ./connected_speech_classification/models/disease_status_classifier.py classify-disease-label \
	--classify_jointly \
    --dataset_dir "your-ad-dataset" \
    --dataset_dir "your-amyloid-dataset"

Formatting the Results

This assumes that the mlflow database has been exported to a .csv file.

python ./connected_speech_classification/evaluation/format_mlflow_results.py convert-mlflow-tables

Combined Script

For convenience, there are scripts in scripts/combined, but they might need some additional changes.