Rational partitioning of spectral feature space for effective clustering of massive spectral image data

Yusei Ito, Yasuo Takeichi, Hideitsu Hino and Kanta Ono

Scientific Reports 14, 22549 (2024). https://doi.org/10.1038/s41598-024-74016-0

Citation

@article{10.1038/s41598-024-74016-0, 
year = {2024}, 
title = {Rational partitioning of spectral feature space for effective clustering of massive spectral image data}, 
author = {Ito, Yusei and Takeichi, Yasuo and Hino, Hideitsu and Ono, Kanta}, 
journal = {Scientific Reports}, 
doi = {10.1038/s41598-024-74016-0}, 
pages = {22549}, 
number = {1}, 
volume = {14}
}

Steps of our clustering method

1. Preparing Voronoi diagram site points.
   The implementation is in the '/MakeVoronoiPoints' directory.

2. Clustering by performing Voronoi tessellation.
   The implementation is in the '/Clustering' directory.

Details of each step are provided in the following sections.

How to prepare Voronoi diagram site points

Use 'main' function in the '/MakeVoronoiPoints/MakeVoronoiPoints.py' file. The parameters are as follows:

• spectra(ndarray): Initial candidate spectra of Voronoi diagram site points. Each column represents a spectrum.
• ref(float): Reference counts (Noise parameter for calculating Eq.(3)).
• mul(float): Constant multiplication parameter(Noise parameter).
• offset(float): Constant sum parameter (Noise parameter).
• generate_num(int): Number of trials of generating spectrum.
• generate_similarity_limit(float): Similarity threshold when generating spectra (Note that this is percent point).
• reduce_similarity_limit(float): Similarity threshold when reducing candidate spectra by hierarchical clustering (Note that this is percent point).
• path_save(str): Path of save file (.csv).

How to perform clustering

Use 'Clustering_by_VoronoiTessellation' function in the '/Clustering/Clustering_VoronoiTessellation.jl' file. This function performs clustering and removing isolation points by using spatial correlation.
Then, use 'main' function in the '/Clustering/Dilate_each_cluster.py' file for complementing the removing points.
The parameters are as follows:

Clustering by VoronoiTessellation

• path_of_experiment_data(str): Path of spectral imaging data (.tif).
• path_of_voronoi_points(str): Path of voronoi points data (.csv).
• kernel(matrix): K_size x K_size (parameter of integrating spatial correlation) matrix of ones.
• kernel_delta(int): Parameter of integrating spatial correlation (kernel_delta = K_size x K_size - P_num)
• path_of_save_folder(str): Path of save folder.

Dilate each cluster

• path_of_save_folder(str): Path of save folder.
• load_folder(str): Path of load folder.

Test code

Preparing test data

Download test data: [figshare] in the same folder.

Preparing Voronoi diagram site points

python MakeVoronoiPoints.py

Clustering by VoronoiTessellation

julia Clustering_VoronoiTessellation.jl

python Dilate_each_cluster.py

Note that the result is also in the test data folder. Running it will result in a FileExistsError, therefore you must either move or delete the resulting folder.