Fig. 3

InstantDL achieves competitive performance on published datasets and computer vision challenges without hyperparameter tuning. a InstantDLs instance segmentation achieves competitive results on the nuclear detection challenge dataset [13], which contains a variety of experimental conditions, cell types, magnifications, and imaging modalities. We show one exemplary image from the dataset and the corresponding prediction using InstantDL’s instance segmentation. The winner of the challenge achieved a Jaccard index of 0.63 (solid line), while the median participant achieved 0.42 (dotted line). InstandDLs instance segmentation achieved a median Jaccard index of 0.60 without hyperparameter tuning. We estimate the Jaccard index distribution by bootstrapping, sampling 100 times half of the test set. Boxes indicate the median and the 25/75%ile of the distribution, whiskers indicate the 1.5 interquartile range. b For the challenge of segmenting nuclei in microscopy images of multiple organs with hematoxylin and eosin staining [14, 15], the winner achieved a Jaccard index of 0.69 (solid line) and the median participant 0.63 (dotted line). InstantDL using instance segmentation reached a Jaccard index of 0.29, and 0.57 using semantic segmentation. c Evaluation of instance segmentation of lung CT images from the Vessel-12 dataset [16]. The winner of the challenge reached an area under the receiver operating characteristic curve (AUC) of 0.99, while the median participant reached 0.94. InstantDL reached an AUC of 0.90 with semantic segmentation, and 0.94 with instance segmentation. d InstantDL’s pixel-wise regression performs similarly well as the published approach ([7] for in-silico staining of bright-field images in three dimensions, but with a higher variability. We achieved a median pearson correlation of 0.85 for nuclear envelope staining and 0.78 for mitochondria staining. e For classification of leukemic blast cell images vs. benign white blood cell images [17, 18], InstantDL achieved an AUC of 0.99, while Matek et al. report 0.99. f Classification of metastatic cancer in small image patches taken from larger digital pathology scans on histopathological images [19]. InstantDL achieved an AUC of 0.93 while the winner of the challenge achieved an AUC of 1.0 and the median participant 0.91