5. Computing Metrics¶

Compute the main reference-free dose metrics and a reference-based geometry comparison using two real aligned studies.

In [1]:

from dosemetrics.metrics import conformity, dvh, homogeneity
from dosemetrics.utils import (
    compare_structure_geometry,
    load_example_study,
    plot_dose_slice,
    plot_metric_values,
    plot_subject_dvhs,
)

dose, structures = load_example_study("longitudinal/time_point_1")
_, later_structures = load_example_study("longitudinal/time_point_2")
ptv = structures["PTV"]

from dosemetrics.metrics import conformity, dvh, homogeneity from dosemetrics.utils import ( compare_structure_geometry, load_example_study, plot_dose_slice, plot_metric_values, plot_subject_dvhs, ) dose, structures = load_example_study("longitudinal/time_point_1") _, later_structures = load_example_study("longitudinal/time_point_2") ptv = structures["PTV"]

DVH metrics¶

In [2]:

dose_bins, volume_percent = dvh.compute_dvh(dose, ptv, verbose=True)
statistics = dvh.compute_dose_statistics(dose, ptv, verbose=True)

plot_dose_slice(
    dose,
    structures=structures,
    structure_names=["PTV", "Brainstem", "Chiasm"],
    contour_colors={"PTV": "white", "Brainstem": "lime", "Chiasm": "cyan"},
    cmap="turbo",
);
plot_subject_dvhs(
    dose,
    structures,
    ["PTV", "Brainstem", "Chiasm", "Eye_L", "Eye_R"],
    color_by_type=False,
);

dose_bins, volume_percent = dvh.compute_dvh(dose, ptv, verbose=True) statistics = dvh.compute_dose_statistics(dose, ptv, verbose=True) plot_dose_slice( dose, structures=structures, structure_names=["PTV", "Brainstem", "Chiasm"], contour_colors={"PTV": "white", "Brainstem": "lime", "Chiasm": "cyan"}, cmap="turbo", ); plot_subject_dvhs( dose, structures, ["PTV", "Brainstem", "Chiasm", "Eye_L", "Eye_R"], color_by_type=False, );

PTV DVH: 659 bins, 0.00-65.80 Gy, 0.0-100.0% volume
PTV dose statistics (Gy)
  Mean    60.15
  Minimum 48.60
  Maximum 65.71
  D98     56.57
  D95     57.66
  D50     60.25
  D05     62.32
  D02     62.81

Conformity and homogeneity¶

In [3]:

prescription = 60.0
quality = {
    "Coverage": conformity.compute_coverage(dose, ptv, prescription),
    "Spillage": conformity.compute_spillage(dose, ptv, prescription),
    "Paddick CI": conformity.compute_paddick_conformity_index(dose, ptv, prescription),
    "Homogeneity index": homogeneity.compute_homogeneity_index(dose, ptv),
    "Dose CV": homogeneity.compute_dose_homogeneity(dose, ptv),
}
plot_metric_values(quality, title="PTV conformity and homogeneity", ylim=(0, 1.1));
quality

prescription = 60.0 quality = { "Coverage": conformity.compute_coverage(dose, ptv, prescription), "Spillage": conformity.compute_spillage(dose, ptv, prescription), "Paddick CI": conformity.compute_paddick_conformity_index(dose, ptv, prescription), "Homogeneity index": homogeneity.compute_homogeneity_index(dose, ptv), "Dose CV": homogeneity.compute_dose_homogeneity(dose, ptv), } plot_metric_values(quality, title="PTV conformity and homogeneity", ylim=(0, 1.1)); quality

Out[3]:

{'Coverage': 0.575361500716488,
 'Spillage': 0.028841572910177006,
 'Paddick CI': 0.5587671700438644,
 'Homogeneity index': 0.10346541102749483,
 'Dose CV': 0.024713383987545967}

Equivalent uniform dose¶

In [4]:

eud = {
    "PTV (a=-10)": dvh.compute_equivalent_uniform_dose(dose, ptv, -10),
    "Brainstem (a=3)": dvh.compute_equivalent_uniform_dose(dose, structures["Brainstem"], 3),
    "Chiasm (a=3)": dvh.compute_equivalent_uniform_dose(dose, structures["Chiasm"], 3),
}
plot_metric_values(eud, title="Equivalent uniform dose", ylabel="Dose (Gy)");
eud

eud = { "PTV (a=-10)": dvh.compute_equivalent_uniform_dose(dose, ptv, -10), "Brainstem (a=3)": dvh.compute_equivalent_uniform_dose(dose, structures["Brainstem"], 3), "Chiasm (a=3)": dvh.compute_equivalent_uniform_dose(dose, structures["Chiasm"], 3), } plot_metric_values(eud, title="Equivalent uniform dose", ylabel="Dose (Gy)"); eud

Out[4]:

{'PTV (a=-10)': 59.922055573464604,
 'Brainstem (a=3)': 23.07795524730258,
 'Chiasm (a=3)': 44.2957090954421}

Reference-based geometric metrics¶

Here the later contours are evaluated against the first time point.

In [5]:

compare_structure_geometry(
    structures,
    later_structures,
    ["PTV", "Brainstem", "Chiasm", "OpticNerve_L"],
).round(3)

compare_structure_geometry( structures, later_structures, ["PTV", "Brainstem", "Chiasm", "OpticNerve_L"], ).round(3)

Out[5]:

	Dice	Jaccard	Volume change (%)
Structure
PTV	0.823	0.699	-13.941
Brainstem	0.801	0.669	-1.967
Chiasm	0.012	0.006	5.063
OpticNerve_L	0.105	0.055	42.857