visualize scan direction

src/depiction_targeted_preproc/pipeline_config/artifacts_mapping.py

@@ -23,6 +23,7 @@
         #"qc/plot_spectra_for_marker.pdf",
         # "qc/plot_sample_spectra_before_after.pdf",
         "qc/plot_peak_counts.pdf",
+        "qc/plot_scan_direction.pdf",
         "cluster_default_kmeans.hdf5",
         "cluster_default_stats_kmeans.csv",
         "cluster_default_kmeans.png",

src/depiction_targeted_preproc/workflow/qc/plot_scan_direction.py

@@ -0,0 +1,82 @@
+# TODO has been migrated/refactored in the new workflow folder
+from pathlib import Path
+from typing import Optional, Annotated
+
+import matplotlib.pyplot as plt
+import numpy as np
+import typer
+from skimage.transform import resize_local_mean
+from typer import Option
+
+from depiction.persistence import ImzmlReadFile
+
+
+def qc_plot_scan_direction(
+    input_imzml_path: Annotated[Path, Option()],
+    output_pdf: Annotated[Path, Option()],
+) -> None:
+    fig = plt.figure(figsize=(10, 10))
+    ax = fig.add_subplot(111)
+    imzml = ImzmlReadFile(input_imzml_path)
+    visualize_coordinates_direction(
+        coordinates=imzml.coordinates_2d,
+        ax=ax,
+        target_n_width=30,
+        target_n_height=30,
+        max_distance=2.0,
+    )
+    fig.savefig(output_pdf)
+
+
+def visualize_coordinates_direction(
+    coordinates: np.ndarray,
+    ax: Optional[plt.Axes] = None,
+    target_n_width: int = 30,
+    target_n_height: int = 30,
+    max_distance: float = 2.0,
+):
+    """
+    Visualizes the scan direction based on the coordinates array.
+    This approach is far from perfect having some important limitations to consider
+    (and also be aware of it's discrepancies from matplotlib's streamplot).
+
+    Limitations:
+    - Uses means in the local neighborhood to determine the direction, i.e. there are cases where the image would
+      be misleading (e.g. frequently alternating directions).
+    - Borders are not handled correctly, since background is mixed in there.
+
+    :param coordinates: (n, 2) or (n, 3) array of coordinates
+    :param ax: matplotlib axis (or None if currently active axis should be used)
+    :param target_n_width: number of points along the width of image to downsample the distance vectorfield to
+    :param target_n_height: number of points along the height of image to downsample the distance vectorfield to
+    :param max_distance: the maximum distance to consider for the vectorfield (before resizing)
+    """
+    ax = ax if ax is not None else plt.gca()
+
+    coordinates = coordinates[:, :2] - coordinates.min(axis=0)[:2]
+    domain = coordinates.max(axis=0)
+    dir_vec = np.diff(coordinates, axis=0)
+    dir_mat_2d = np.zeros((domain[0] + 1, domain[1] + 1, 2))
+    dir_mat_2d[coordinates[:, 0], coordinates[:, 1], :] = np.concatenate([dir_vec, [[0, 0]]], axis=0)
+
+    thresholded_dir_mat_2d = dir_mat_2d.copy()
+    thresholded_dir_mat_2d[np.where(np.linalg.norm(dir_mat_2d, axis=2) > max_distance)] = 0
+
+    thresholded_dir_img = resize_local_mean(thresholded_dir_mat_2d, (target_n_width, target_n_height), grid_mode=False)
+
+    arrow_locations = np.argwhere(thresholded_dir_img.sum(axis=2) != 0)
+    arrow_directions = thresholded_dir_img[arrow_locations[:, 0], arrow_locations[:, 1], :]
+
+    ax.quiver(
+        arrow_locations[:, 0],
+        arrow_locations[:, 1],
+        arrow_directions[:, 0],
+        arrow_directions[:, 1],
+        scale=1,
+        units="xy",
+    )
+    ax.invert_yaxis()
+
+
+if __name__ == "__main__":
+    typer.run(qc_plot_scan_direction)

src/depiction_targeted_preproc/workflow/rules/rules_qc.smk

@@ -133,3 +133,13 @@ rule qc_plot_peak_counts:
         "python -m depiction_targeted_preproc.workflow.qc.plot_peak_counts"
         " --config-path {input.config} --imzml-peaks {input.imzml[0]}"
         " --output-pdf {output.pdf}"
+
+rule qc_plot_scan_direction:
+    input:
+        imzml=multiext("{sample}/corrected.peaks",".imzML",".ibd"),
+    output:
+        pdf="{sample}/qc/plot_scan_direction.pdf"
+    shell:
+        "python -m depiction_targeted_preproc.workflow.qc.plot_scan_direction"
+        " --input-imzml-path {input.imzml[0]}"
+        " --output-pdf {output.pdf}"