Extended Data Fig. 4: Association of PET/CT image properties with evaluation metrics. | Nature Machine Intelligence

Extended Data Fig. 4: Association of PET/CT image properties with evaluation metrics.

From: Results from the autoPET challenge on fully automated lesion segmentation in oncologic PET/CT imaging

Extended Data Fig. 4

Left column: Mixed effects analysis revealed a positive association between the Dice score and the mean lesion volume (p = 0.0002) and the Dice score and the lesion count (p = 0.008). Middle column: Mixed effects analysis revealed a positive association between the BMI and the false positive volume (p = 0.006) and a negative association between the lesion count and the false positive volume (p = 0.008). Right Column: Mixed effect analysis revealed a negative association between mean lesion volume and false negative volume (p = 0.01). These findings indicate that the detection of small lesions was challenging for submitted algorithms and that patient-related factors such as BMI may have an impact on overall algorithm performance. Mixed effects analysis was performed in Python using the Statsmodels module (Version 0.14.0) based on a two-tailed Wald test. P-values are corrected for multiple testing (Bonferroni correction).

Back to article page