anatomical_metrics.py

from numbers import Real
from typing import Literal, Mapping

import pandas as pd

from vital.metrics.evaluate.segmentation import check_metric_validity
from vital.results.metrics import Metrics


class AnatomicalMetrics(Metrics):
    """Abstract class that computes anatomical metrics on the results and saves them to csv."""

    desc = "anatomical_metrics"
    thresholds: Mapping[str, Mapping[Literal["min", "max"], Real]]  #: Anatomical metrics' threshold values

    @classmethod
    def _aggregate_metrics(cls, metrics: pd.DataFrame) -> pd.DataFrame:
        """Computes the number of results that were anatomically invalid for each anatomical metric.

        Args:
            metrics: Metrics computed over each result.

        Returns:
            Number of results that were anatomically invalid for each anatomical metric.
        """

        def count_metric_errors(metric_name):
            return lambda series: sum(
                not check_metric_validity(
                    metric_value, thresholds=cls.thresholds.get(metric_name), optional_structure=False
                )
                for metric_value in series
            )

        aggregation_dict = {metric_name: count_metric_errors(metric_name) for metric_name in metrics.keys()}
        aggregated_metrics = metrics.agg(aggregation_dict)
        aggregated_metrics.name = "anatomical_errors_count"  # The series' name will be its index in the dataframe
        return pd.DataFrame(aggregated_metrics).T