Multi-stain deep learning prediction model of treatment response in lupus nephritis based on renal histopathology

Cheng Cheng ¹, Bin Li ², Jie Li ¹

¹Department of Pediatrics, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.
²Clinical Trials Unit, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.
³Department of Nephrology, The First Affiliated Hospital, Sun Yat-Sen University, Guangzhou, Guangdong, China; National Health Commission Key Laboratory of Clinical Nephrology (Sun Yat-sen University) and Guangdong Provincial Key Laboratory of Nephrology, Guangzhou, Guangdong, China.
⁴Department of Medical Ultrasonics, Institute of Diagnostic and Interventional Ultrasound, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.
⁵Department of Nephrology, Zhongshan City People's Hospital, Zhongshan, Guangdong, China.
⁶Department of Pediatrics, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.
⁷Department of Nephrology, Guangzhou First People's Hospital, Guangzhou, Guangdong, China.
⁸Department of Pediatrics, Guangzhou First People's Hospital, Guangzhou, Guangdong, China.
⁹Department of Pediatrics, Zhongshan City People's Hospital, Zhongshan, Guangdong, China.
¹⁰Clinical Trials Unit, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China; Institute of Precision Medicine, the First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China; Department of Gastroenterology and Hepatology, the First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.

⁰Department of Pediatrics, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China.

Kidney International +

|

December 29, 2024

View abstract on PubMed

Summary

Deep learning models applied to kidney biopsies can predict treatment response in lupus nephritis patients. This approach may aid in risk stratification and therapeutic decisions for better patient outcomes.

Area Of Science

Nephrology
Pathology
Artificial Intelligence

Background

Predicting treatment response in lupus nephritis is crucial for prognosis but lacks effective tools.
Kidney biopsy analysis is key to understanding disease progression and treatment outcomes.

Purpose Of The Study

To develop and validate deep learning models for predicting 12-month treatment response in lupus nephritis using kidney biopsies.
To assess the performance of these models against conventional clinicopathologic parameters.

Main Methods

Deep learning models were trained on digital slides from hematoxylin and eosin-, periodic acid-Schiff-, periodic Schiff-methenamine silver-, and Masson's trichrome-stained kidney biopsies.
Models were developed using a cohort of 245 patients and validated on an external cohort of 71 patients.
The primary outcome was complete treatment response at 12 months, defined by proteinuria and estimated glomerular filtration rate.

Main Results

Single-stain deep learning models achieved areas under the curve (AUC) ranging from 0.813 to 0.862.
An integrated multi-stain model reached an AUC of 0.901 (internal validation) and 0.840 (external testing).
The multi-stain model outperformed traditional clinicopathologic parameters in predicting treatment response.

Conclusions

Deep learning models utilizing kidney biopsy data show promise for predicting treatment response in lupus nephritis.
Identified key predictive features include tertiary lymphoid structures, glomerulosclerosis, interstitial fibrosis, and tubular atrophy.
Further validation is needed for clinical implementation in risk stratification and therapeutic decision-making.

Keywords:

artificial intelligence kidney pathology lupus nephritis prediction model treatment response