Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Feb 26;25(1):66.
doi: 10.1186/s12880-025-01606-3.

Using deep learning to differentiate among histology renal tumor types in computed tomography scans

Hung-Cheng Kan  1   2   3 , Po-Hung Lin  2   3   4 , I-Hung Shao  2   3   4 , Shih-Chun Cheng  5 , Tzuo-Yau Fan  6 , Ying-Hsu Chang  3   7 , Liang-Kang Huang  2   3 , Yuan-Cheng Chu  2   3 , Kai-Jie Yu  2   3 , Cheng-Keng Chuang  2   3 , Chun-Te Wu  2   3 , See-Tong Pang  2   3 , Syu-Jyun Peng  8   9
Affiliations

Using deep learning to differentiate among histology renal tumor types in computed tomography scans

Hung-Cheng Kan et al. BMC Med Imaging. .

Abstract

Background: This study employed a convolutional neural network (CNN) to analyze computed tomography (CT) scans with the aim of differentiating among renal tumors according to histologic sub-type.

Methods: Contrast-enhanced CT images were collected from patients with renal tumors. The patient cohort was randomly split to create a training dataset (90%) and a testing dataset (10%). Following image dataset augmentation, Inception V3 and Resnet50 models were used to differentiate between renal tumors subtypes, including angiomyolipoma (AML), oncocytoma, clear cell renal cell carcinoma (ccRCC), chromophobe renal cell carcinoma (chRCC), and papillary renal cell carcinoma (pRCC). 5-fold cross validation was then used to evaluate the models in terms of classification performance.

Results: The study cohort comprised 554 patients, including those with angiomyolipoma (n = 67), oncocytoma (n = 34), clear cell renal cell carcinoma (n = 246), chromophobe renal cell carcinoma (n = 124), and papillary renal cell carcinoma (n = 83). Dataset augmentation of the training dataset included this to 4238 CT images for analysis. The accuracy of the models was as follows: Inception V3 (0.830) and Resnet 50 (0.849).

Conclusion: This study demonstrated the efficacy of using deep learning models for the classification of renal tumor subtypes from contrast-enhanced CT images. While the models showed promising accuracy, further development is necessary to improve their clinical applicability.

Keywords: Artificial intelligence; Deep learning; Histology; Renal tumor.

PubMed Disclaimer

Conflict of interest statement

Declarations. Ethics approval and consent to participate: This study was approved by the ethics committee at Chang Gung Medical Foundation, Taiwan. After reviewing the project application, due to the retrospective nature of this study, the necessity of individual patient informed consent was waived by the ethics committee at Chang Gung Medical Foundation. All methods and procedures in this study were conducted in accordance with relevant guidelines and regulations. Consent for publication: Not applicable. Competing interests: The authors declare no competing interests. Clinical trial number: Not applicable.

Figures

Fig. 1
Fig. 1
Flowchart showing patient inclusion and exclusion criteria
Fig. 2
Fig. 2
Image preprocessing and normalization process
Fig. 3
Fig. 3
Results of Inception V3: (A) Accuracy; (B) Weighted precision; (C) Macro F1-score; (D) Weighted F1-score. The highest accuracy (0.804 ± 0.019), the highest precision (0.847 ± 0.021), and the highest F1-score (0.813 ± 0.0176) were all achieved using 220 trainable layers
Fig. 4
Fig. 4
Results of Resnet 50: (A) Accuracy; (B) Weighted precision; (C) Macro F1-score; (D) Weighted F1-score. The highest average accuracy (0.811 ± 0.027) was achieved using 50 trainable layers. The highest average weighted precision (0.865 ± 0.015) was achieved using 150 trainable layers. The highest average macro F1-score (0.753 ± 0.040) was achieved using 75 trainable layers. The highest average weighted F1-score (0.838 ± 0.027) was achieved using 50 trainable layers
See this image and copyright information in PMC

References

    1. Fateh SM, Arkawazi LA, Tahir SH, Rashid RJ, Rahman DH, Aghaways I, Kakamad FH, Salih AM, Bapir R, Fakhralddin SS, Fattah FH, Abdalla BA, Mohammed SH. Renal cell carcinoma T staging: diagnostic accuracy of preoperative contrast-enhanced computed tomography. Mol Clin Oncol. 2023;18(2):11. - PMC - PubMed
    1. Johnson DC, Vukina J, Smith AB, et al. Preoperatively misclassified, surgically removed benign renal masses: a systematic review of surgical series and United States population level burden estimate. J Urol. 2015;193:30. - PubMed
    1. Kutikov A, Fossett LK, Ramchandani P, et al. Incidence of benign pathologic findings at partial nephrectomy for solitary renal mass presumed to be renal cell carcinoma on preoperative imaging. Urology. 2006;68:737. - PubMed
    1. Patel HD, Johnson MH, Pierorazio PM, Sozio SM, Sharma R, Iyoha E, Bass EB, Allaf ME. Diagnostic accuracy and risks of biopsy in the diagnosis of a renal mass suspicious for localized renal cell carcinoma: systematic review of the literature. J Urol. 2016;195(5):1340–7. - PMC - PubMed
    1. Prince J, Bultman E, Hinshaw L, et al. Patient and tumor characteristics can predict nondiagnostic renal mass biopsy findings. J Urol. 2015;193:1899. - PMC - PubMed

MeSH terms