Automated acquisition of explainable knowledge from unannotated histopathology images

Abstract

Deep learning algorithms have been successfully used in medical image classification. In the next stage, the technology of acquiring explainable knowledge from medical images is highly desired. Here we show that deep learning algorithm enables automated acquisition of explainable features from diagnostic annotation-free histopathology images. We compare the prediction accuracy of prostate cancer recurrence using our algorithm-generated features with that of diagnosis by expert pathologists using established criteria on 13,188 whole-mount pathology images consisting of over 86 billion image patches. Our method not only reveals findings established by humans but also features that have not been recognized, showing higher accuracy than human in prognostic prediction. Combining both our algorithm-generated features and human-established criteria predicts the recurrence more accurately than using either method alone. We confirm robustness of our method using external validation datasets including 2276 pathology images. This study opens up fields of machine learning analysis for discovering uncharted knowledge.

Development and interpretation of a pathomics-based model for the prediction of microsatellite instability in Colorectal Cancer

“…With the continuous penetration of artificial intelligence (AI) into the field of medical imaging, researchers have sought solutions based on deep learning, a research area in AI, in a wide range of medical problems, such as prediction of gene mutations 14 and tumor-infiltrating lymphocytes 12 , and cancer screening 15 , 16 . Whereas traditional machine learning depends largely on human-selected features 17 , deep learning can learn features from the data, which makes it possible for researchers to discover untapped information 18 , 19 . Previous studies have suggested that deep learning can discover regions that contribute to microsatellite (MS) status with special pathomorphological characteristics 20 , but the applicability of the model in the Asian population remains in question because of the great variation in demographics and data preparation.…”

The potential of convolutional neural network diagnosing prostate cancer

“…The evaluation performed by Yamamoto et al consisted of comparing the information generated by his method with the Gleason score, reaching the AUC of 0.82 for the prediction of the biochemical recurrence of prostate cancer. Despite the excellent accuracy, the objective of our proposed work differs from Yamamoto's (Yamamoto et al 2019), since we aim to classify tissues. Although these studies presented good results, there are a variety of limitations regarding application, potential of accuracy, and number of images.…”

“…Even though reasonable, the presented results still have a wide margin of accuracy to be improved. Yamamoto,et al used 13,188 hole-mount pathology images and his method consisted of implementing deep learning networks with auto encoders in order to predict the biochemical recurrence of prostate cancer (Yamamoto et al 2019).…”

Facial UV photo imaging for skin pigmentation assessment using conditional generative adversarial networks

“…The recent development of deep learning technologies enables highly accurate image analysis in various fields. Remarkable progresses have been achieved particularly in medical image analysis by the deep learning technologies such as lesion detection in x-ray images, histopathological image analysis, and disease name classification for clinical photo images 5 – 11 . The increase of both computational resources and manually curated annotation data enables deep learning-based methods to lead accurate diagnosis results.…”