Fig. 2: Pipeline overview of MIXTURE. | Modern Pathology

Fig. 2: Pipeline overview of MIXTURE.

From: MIXTURE of human expertise and deep learning—developing an explainable model for predicting pathological diagnosis and survival in patients with interstitial lung disease

Fig. 2

For each magnification, elemental feature extractors (ElEx) were trained using self-supervised learning. This feature extractor consists of a ResNet18 CNN which outputs features consisting of 128 vectors. The extracted features were clustered throughout the principal pretraining set. The pathologists viewed a montage of each cluster tiles and reclassified them into pathologically meaningful findings. Finally, the reclassified findings were used as labels of training data for the transfer learning of feature extractor to obtain a classifier to classify the findings from the tiles.

Back to article page