iSegMSI provides a novel interactive segmentation strategy for MSI data. It could improve segmentation results by subdividing or merging inappropriate regions into proper regions specified by scribbles. The proposed iSegMSI model framework supports three dimensionality reduction methods, PCA, PCA + t-SNE and UMAP. In pratical, a researcher can choose different dimensionality reduction algorithms and any dimensions for downstream data analysis or retain higher dimensions to improve segmentation performance. Developer is Lei Guo from Laboratory of Biomedical Network, Department of Electronic Science, Xiamen University of China.
Schematic overflow of the iSegMSI model. (A) Architecture of the iSegMSI model. The method consists of two modules, including a dimensionality reduction (DR) module, and a feature clustering (FC) module that is consisted of a CNN block and an argmax classifier. (B) CNN block. Each of the first N-1 CNN components contain a 2D convolutional layer (p filters with 3×3 kernel size), a batch normalization, and a ReLU layer. The last component contains a 2D convolutional layer (q filters with 1×1 kernel size) and a batch normalization layer.
python == 3.5, 3.6 or 3.7
pytorch == 1.8.2
opencv == 4.5.3
matplotlib == 2.2.2
numpy >= 1.8.0
umap == 0.5.1
The input is the preprocessed MSI data with two-dimensional shape [X*Y,P], where X and Y represent the pixel numbers of horizontal and vertical coordinates of MSI data, and P represents the number of ions.
cd to the iSegMSI fold
If you want to perfrom iSegMSI for unsupervised segmentation, taking fetus mouse data as an example, run:
python run.py -input_file .../data/fetus_mouse.txt --input_shape 202 107 1237 --DR_mode umap --n_components 3 --use_scribble 0 --output_file output.txt
If you want to perfrom iSegMSI for interactive segmentation, taking fetus mouse data as an example, run:
python run.py -input_file .../data/fetus_mouse.txt --input_shape 202 107 1237 --DR_mode umap --n_components 3 --use_scribble 1 -- input_scribble .../data/fetus_mouse_scribble.txt --output_file output.txt
If you want to perfrom iSegMSI for hyperparameter search, taking fetus mouse data as an example, run:
python hyperparameter_earch.py -input_file .../data/fetus_mouse.txt --input_shape 202 107 1237 --DR_mode umap --n_components 3 --use_scribble 0 --output_file output
Please contact me if you have any help: gl5121405@gmail.com
Lei Guo, Xingxing Liu, et al. iSegMSI: An Interactive Strategy to Improve Spatial Segmentation of Mass Spectrometry Imaging Data. Analytical Chemistry, 2022, 94, 42, 14522–14529.