PICASO Set Operator for Computational Nephropathology.

IF 3.2 Q1 UROLOGY & NEPHROLOGY Kidney360 Pub Date : 2025-03-03 DOI:10.34067/KID.0000000668

Samira Zare, Huy Q Vo, Nicola Altini, Vitoantonio Bevilacqua, Michele Rossini, Francesco Pesce, Loreto Gesualdo, Sándor Turkevi-Nagy, Jan Ulrich Becker, Chandra Mohan, Hien Van Nguyen

{"title":"PICASO Set Operator for Computational Nephropathology.","authors":"Samira Zare, Huy Q Vo, Nicola Altini, Vitoantonio Bevilacqua, Michele Rossini, Francesco Pesce, Loreto Gesualdo, Sándor Turkevi-Nagy, Jan Ulrich Becker, Chandra Mohan, Hien Van Nguyen","doi":"10.34067/KID.0000000668","DOIUrl":null,"url":null,"abstract":"Background: The advent of digital nephropathology offers the potential to integrate deep learning algorithms into the diagnostic workflow. We introduce PICASO, a novel permutation-invariant set operator to dynamically aggregate histopathologic features from instances. We applied PICASO to two nephropathology scenarios: detecting active crescent lesions in sets of glomerular crops with IgA nephropathy and case-level classification for antibody-mediated rejection (AMR) in kidney transplant.Methods: PICASO is a Transformer-based set operator that aggregates features from sets of instances to make predictions. It utilizes initial Histopathologic Vectors as a static memory component and continuously updates them based on input embeddings. For active crescent detection in IgA nephropathy cases, we obtained 6206 Periodic acid-Schiff-stained (PAS) glomerular crops (5792 no Active Crescent, 414 Active Crescent) from three different health institutes. For the AMR classification, we have 1655 PAS glomerular crops (769 AMR and 886 Non-AMR images) from 89 biopsies. The performance of PICASO as a set operator was compared with other set operators such as DeepSet, Set Transformer, DeepSet++, and Set Transformer++ using metrics including area under the receiver operating characteristic curves (AUROC), area under the precision-recall curves (AUPR), recall, and accuracy.Results: PICASO achieved superior performance in detecting active crescent in IgA nephropathy cases, with an AUROC of 0.99 (95% confidence interval, 0.98 to 0.99) on internal validation and 0.96 (95% confidence interval, 0.95 to 0.98) on external validation, significantly outperforming other set operators (P<0.001). It also attained the highest AUROC of 0.97 (95% confidence interval, 0.90 to 1.0, P=0.02) for case-level AMR classification. The AUPR, recall, and accuracy scores were also higher when using PICASO, and it significantly outperformed baselines (P<0.001).Conclusions: PICASO can potentially advance nephropathology by improving performance through dynamic feature aggregation.","PeriodicalId":17882,"journal":{"name":"Kidney360","volume":" ","pages":""},"PeriodicalIF":3.2000,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Kidney360","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.34067/KID.0000000668","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"UROLOGY & NEPHROLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Background: The advent of digital nephropathology offers the potential to integrate deep learning algorithms into the diagnostic workflow. We introduce PICASO, a novel permutation-invariant set operator to dynamically aggregate histopathologic features from instances. We applied PICASO to two nephropathology scenarios: detecting active crescent lesions in sets of glomerular crops with IgA nephropathy and case-level classification for antibody-mediated rejection (AMR) in kidney transplant.

Methods: PICASO is a Transformer-based set operator that aggregates features from sets of instances to make predictions. It utilizes initial Histopathologic Vectors as a static memory component and continuously updates them based on input embeddings. For active crescent detection in IgA nephropathy cases, we obtained 6206 Periodic acid-Schiff-stained (PAS) glomerular crops (5792 no Active Crescent, 414 Active Crescent) from three different health institutes. For the AMR classification, we have 1655 PAS glomerular crops (769 AMR and 886 Non-AMR images) from 89 biopsies. The performance of PICASO as a set operator was compared with other set operators such as DeepSet, Set Transformer, DeepSet++, and Set Transformer++ using metrics including area under the receiver operating characteristic curves (AUROC), area under the precision-recall curves (AUPR), recall, and accuracy.

Results: PICASO achieved superior performance in detecting active crescent in IgA nephropathy cases, with an AUROC of 0.99 (95% confidence interval, 0.98 to 0.99) on internal validation and 0.96 (95% confidence interval, 0.95 to 0.98) on external validation, significantly outperforming other set operators (P<0.001). It also attained the highest AUROC of 0.97 (95% confidence interval, 0.90 to 1.0, P=0.02) for case-level AMR classification. The AUPR, recall, and accuracy scores were also higher when using PICASO, and it significantly outperformed baselines (P<0.001).

Conclusions: PICASO can potentially advance nephropathology by improving performance through dynamic feature aggregation.

查看原文