Unsupervised machine learning for identifying important visual features through bag-of-words using histopathology data from chronic kidney disease

Abstract

Pathologists use visual classification to assess patient kidney biopsy samples when diagnosing the underlying cause of kidney disease. However, the assessment is qualitative, or semi-quantitative at best, and reproducibility is challenging. To discover previously unknown features which predict patient outcomes and overcome substantial interobserver variability, we developed an unsupervised bag-of-words model. Our study applied to the C-PROBE cohort of patients with chronic kidney disease (CKD). 107,471 histopathology images were obtained from 161 biopsy cores and identified important morphological features in biopsy tissue that are highly predictive of the presence of CKD both at the time of biopsy and in one year. To evaluate the performance of our model, we estimated the AUC and its 95% confidence interval. We show that this method is reliable and reproducible and can achieve 0.93 AUC at predicting glomerular filtration rate at the time of biopsy as well as predicting a loss of function at one year. Additionally, with this method, we ranked the identified morphological features according to their importance as diagnostic markers for chronic kidney disease. In this study, we have demonstrated the feasibility of using an unsupervised machine learning method without human input in order to predict the level of kidney function in CKD. The results from our study indicate that the visual dictionary, or visual image pattern, obtained from unsupervised machine learning can predict outcomes using machine-derived values that correspond to both known and unknown clinically relevant features.

CluSA: Clustering-based Spatial Analysis framework through Graph Neural Network for Chronic Kidney Disease Prediction using Histopathology Images

“…Previously, we have shown that unsupervised machine-learned clustering features are potential surrogates of predicting eGFR and can be used as tools for prognosis as well as for objective assessment of the level of kidney function in CKD 45 . In the present study, our results demonstrate that the addition of spatial information improves the model's performance by 2.4% and 5.1% of AUC at the biopsy and one-year prediction, respectively, compared to the previous study.…”

“…In order to cluster image patterns on image patches, we extracted features from each image patch for the clustering using pre-trained DeepLab V3+ with ResNet-18 model. 38,45 Then, all 172 biopsy cores on images were tiled into 107,471 patches. Then, those patches were clustered through K-means clustering to group similar image patterns together (Figure 3a).…”

Artificial Intelligence-Assisted Renal Pathology: Advances and Prospects

“…Based on the comprehensive morphological features extracted from 161 renal biopsies, along with patient clinical information, the AUC of the prediction model for eGFR at biopsy time reached 0.93, while that for 1-year eGFR was 0.80. These results indicated the potential of visual-feature-based algorithms for predicting CKD progression [ 113 ].…”

Unsupervised Learning Based on Multiple Descriptors for WSIs Diagnosis

“…The challenges in deep learning for cancer or disease diagnosis on histopathological images include the use of large amounts of accurately annotated data and the generalization limitations of WSIs due to multiple variants and the intricate cellular structure of histopathology data. Publicly available WSIs with accurate annotation or label information are fewer as compared to natural image datasets because acquiring WSIs is a difficult and costly task owing to the specific nature of medical imaging modalities [ 2 , 3 , 4 , 5 ]. To address this issue, previous researchers used pre-trained models based on previously acquired knowledge from different domain image datasets and then applied them to the medical imaging domain.…”