Artificial Intelligence

Scientific analysis evaluates kidney fibrosis diagnosis using smart ultrasound

BUZZINGA team5 min read

A systematic review and rigorous meta-analysis in Scientific Reports evaluated the accuracy of AI-supported ultrasound in detecting moderate-to-severe interstitial fibrosis and tubular atrophy in chronic kidney disease patients. The review included 14 studies with a total of 3,776 participants, extracting pooled performance indicators from 1,244 patients.

The meta-analysis published in Scientific Reports provides a precise quantitative summary of the potential and limitations of using AI-supported ultrasound as a non-invasive tool to assess kidney tissue fibrosis.

Taipei Medical University systematic review scope across 14 studies

A systematic review and rigorous meta-analysis published in Scientific Reports on November 16, 2026, evaluated the efficacy of AI-supported ultrasound imaging in identifying moderate-to-severe cases of interstitial fibrosis and tubular atrophy in individuals with chronic kidney disease.

Researchers at Taipei Medical University and collaborating colleagues examined scientific evidence drawn from 14 research studies comprising a grand total of 3,776 participants, based on an extensive search of international scientific databases extending through October 2025. This makes the paper a novel scientific synthesis of previous primary studies rather than a new clinical trial or newly approved medical device.

Importance of non-invasive screening compared to the reference kidney biopsy

The terms interstitial fibrosis and tubular atrophy describe the chronic structural damage that affects kidney tissue and impacts its vital function. Surgical kidney biopsy is considered the most accurate reference method to assess these histological changes, yet it is an invasive medical procedure carrying risks and cannot be easily repeated for routine follow-up.

In contrast, traditional ultrasound imaging and shear wave elastography provide non-invasive imaging measurements without the need to take a tissue sample. Machine learning systems attempt to interpret these measurements to identify patients most likely to have substantial tissue damage. The review focused on moderate-to-severe changes defined by involvement of at least 25% of the relevant tissue according to the criteria of studies that used biopsy as the reference standard.

Sensitivity, specificity, and pooled diagnostic performance curve results

The authors noted an important statistical distinction: only 1,244 participants contributed to the pooled diagnostic performance estimates out of the total 3,776 participants across the included studies.

Across the pooled data, researchers estimated the screening sensitivity at approximately 0.821 and specificity at approximately 0.828. Sensitivity measures the proportion of those with the condition who were correctly identified, while specificity measures the proportion of those without the condition who were accurately ruled out. Furthermore, the pooled area under the receiver operating characteristic curve reached 0.880 within a 95% confidence interval ranging from 0.815 to 0.924.

Exploratory findings of deep learning and shear wave elastography

These figures represent pooled research findings rather than guaranteed performance for every individual hospital or patient. Diagnostic accuracy may vary depending on disease prevalence, the type of ultrasound equipment, image acquisition methods, algorithmic model structure, and patient clinical characteristics. Moreover, false positives and false negatives retain significant medical impact.

Shear wave elastography techniques and deep learning methods showed numerically favorable performance in certain subgroup analyses, but the authors described these subgroup findings as exploratory only due to the small number of studies available for decisive comparisons and the wide range of statistical uncertainty.

Clinical controls and requirements for future multicenter validation

These studies hold promising importance for initial screening and monitoring processes; however, artificial intelligence output should not automatically replace evaluation by a specialist physician, laboratory tests, or kidney biopsy when clinical conditions warrant them. Reliable application will require prospective multicenter studies across diverse patient cohorts, standardizing reference standards and clinical decision thresholds, and measuring actual impact on patient health.

This systematic review published in Scientific Reports illustrates how medical AI indicators should be read in their proper clinical context. Although recording a sensitivity of 0.821, a specificity of 0.828, and an area under the curve of 0.880 in 1,244 patients demonstrates promising potential to support non-invasive kidney fibrosis screening, the variability in ultrasound devices and patient characteristics necessitates treating these technologies as supportive decision-making tools rather than a complete replacement for biopsy and specialist diagnosis.

Furthermore, future multicenter studies will remain essential to standardize imaging protocols and clinical diagnostic thresholds prior to widespread routine adoption of these models within hospitals.

Frequently asked questions

Question: What is the number of studies and participants in the meta-analysis for kidney fibrosis diagnosis?
Answer: The systematic review included 14 studies with a total of 3,776 participants, while the pooled diagnostic performance estimates were based on data from 1,244 participants who underwent comparison with the reference biopsy.

Question: What levels of accuracy did AI-supported ultrasound demonstrate?
Answer: Pooled sensitivity reached 0.821 and pooled specificity reached 0.828, while the area under the diagnostic performance curve recorded 0.880 with a confidence interval extending from 0.815 to 0.924.

Question: Why are doctors looking for supportive alternatives to traditional kidney biopsy?
Answer: Because biopsy is an invasive surgical procedure that is difficult to repeat regularly to monitor the progression of interstitial fibrosis and tubular atrophy, whereas ultrasound provides a safe and non-invasive examination.

Question: Do imaging artificial intelligence systems currently replace biopsy and specialist evaluation?
Answer: No, the researchers emphasize that the figures represent pooled research estimates influenced by device type and patient characteristics, and they do not replace laboratory tests or biopsy when medically necessary.