Skip to main content
Journal of Big Data
Download PDF
Download ePub

Automatic identification and classification of pediatric glomerulonephritis on ultrasound images based on deep learning and radiomics

Journal of Big Data volume 11, Article number: 165 (2024) Cite this article

Abstract

Background

Glomerulonephritis (GN) encompasses a heterogeneous group of kidney diseases, often presenting with subclinical manifestations in children, leading to frequent missed diagnoses. Renal biopsy, while considered the gold standard, is invasive, prone to sampling errors, and time-consuming, thus hindering rapid diagnosis. This study aimed to develop a noninvasive diagnostic model for childhood GN using renal ultrasound images through the integration of deep learning and radiomics techniques.

Methods

Ultrasound images were acquired from children undergoing ultrasound-guided biopsy. A total of 469 renal ultrasound images were selected and divided into training and validation sets at a ratio of 8:2 to train a U-Net model for precise kidney image segmentation. Using radiomics, a comprehensive set of radiomic features were extracted from the segmented kidney regions. The extracted features were categorized based on GN types: IgA nephropathy (127 cases), minimal change disease (83 cases), and Henoch–Schönlein purpura nephritis (103 cases). These categories were further randomly split into training and validation sets at a ratio of 8:2. Within the training set, analysis of variance (ANOVA) was used for feature selection, followed by supervised Least Absolute Shrinkage and Selection Operator (LASSO) regression for dimensionality reduction, resulting in the selection of 37 features. These features were then integrated with a random forest algorithm to develop a GN classification model. The model's performance was comprehensively evaluated using the validation set.

Results

The segmentation model exhibited remarkable performance during training, achieving an accuracy of 95.19% in the validation set. Thirty-seven features were identified through feature selection, leading to the development of a robust classification model. Evaluation on the validation set revealed high accuracy and predictive power across different GN categories, with Area Under the Curve (AUC) values ranging from 0.91 to 0.98.

Conclusions

The combined use of deep learning and radiomics techniques utilizing renal ultrasound images demonstrates significant potential for classifying childhood GN subtypes. This noninvasive approach holds promise for improving diagnostic efficiency and patient outcomes in GN.

Background

Glomerulonephritis (GN) is a complex and prevalent disease within the spectrum of kidney diseases [1]. Subtypes of GN include IgA nephropathy (IgAN), minimal change disease (MCD), Henoch–Schönlein purpura nephritis (HSPN), membranous nephropathy (MN), focal segmental glomerulosclerosis (FSGS), and membranoproliferative glomerulonephritis (MPGN) [2, 3]. Each subtype exhibits distinct pathological mechanisms and clinical manifestations [4]. Early symptoms of GN in children can be insidious. Non-specific manifestations such as fatigue, edema, and changes in urine output may pose a considerable risk of under-recognition or misdiagnosis [5]. Renal biopsy is the gold standard for diagnosis of GN, but this procedure carries surgical risks, including bleeding, infection, and other complications [6,7,8]. Additionally, it is highly affected by sampling errors [9]. The burdensome and time-consuming nature hinders rapid clinical diagnosis, which may delay treatment and affect the disease prognosis. Now, a range of advanced, non-invasive technologies, including super microvascular imaging, is showing promise in differentiating cases of IgAN from non-IgAN [10]. Additionally, renal ultrasonography radiomics has proven to be effective, to a certain extent, in distinguishing between IgAN and MN [11]. Furthermore, gut microbiome analysis has offered novel insights into the identification of MN [12]. These studies have predominantly focused on adult populations with specific types of GN, neglecting the physiological differences in renal function between children and adults. Moreover, their diagnostic efficacy and clinical feasibility have not yet reached the point where they could stand in for renal biopsy as a gold standard. Therefore, continuous exploration of rapid, accurate, and non-invasive diagnostic methods for GN in children remains essential for improving diagnostic efficiency and improving disease prognosis.

As the growing reliance upon medical imaging technology proceeds apace, the importance of ultrasound imaging in renal disease diagnosis is increasingly recognized [13]. As a non-invasive and accessible technique, ultrasound can provide vital information for initial kidney anatomical and functional assessment [14, 15]. However, precise diagnosis of GN via renal ultrasound imaging is challenging because of difficulties in distinguishing subtle kidney structures and lesion features [16], limiting the diagnostic utility. Recently, radiomics research has shown significant growth, integrating knowledge from radiology, computer science, and data analysis to extract quantitative features from medical images [17]. This approach reveals microstructural and functional abnormalities, exploring disease onset, development, and evolution [18, 19], and thereby providing critical support in early screening and diagnosis, personalized treatment planning, and efficacy assessment [20].

Radiomics research often relies on manual delineation of regions of interest (ROIs), which is inefficient, labor-intensive, and subjective, potentially leading to unstable outcomes [21]. However, recent advancements in deep learning have greatly improved medical image processing, particularly in image segmentation [22]. Multiple algorithms have been employed for this task, demonstrating excellent performance in extracting target regions from images [23, 24]. For instance, U-Net [25], a widely adopted deep learning architecture, and Generative Adversarial Networks [26] have been widely shown to excel in improving image quality and segmentation accuracy. The objective of this study is to integrate deep learning with radiomics techniques to achieve automatic and accurate segmentation and feature extraction from pediatric renal ultrasound images. By utilizing deep learning models, precise segmentation is achieved, and radiomics methods are employed to extract rich image features from the segmented area. These features are then used to establish a diagnostic classification model for GN in children and evaluate the model’s performance.

Method

Patients

The study design was approved by the Ethics Committee of Children’s Hospital at Chongqing Medical University. The study enrolled pediatric patients who underwent ultrasound-guided needle biopsy at our hospital from July 2023 to September 2024. Prior to their inclusion, informed consent was obtained from both the children and their guardians. Inclusion criteria were: (1) availability of pre-renal biopsy renal ultrasound images for direct correlation with pathological status; (2) high-quality ultrasound images clearly depicting renal structures including cortex, medulla, and renal pelvis; (3) availability of comprehensive ultrasound data, renal biopsy results, and clinical information. Cases were excluded if: (1) ultrasound image quality was suboptimal due to technical or operational limitations; (2) ultrasound data, biopsy results, or clinical information were missing; (3) patients were unable to cooperate for a complete ultrasound or had contraindications to renal biopsy.

Ultrasound image acquisition

All ultrasound images were acquired by experienced senior sonographers using the same model of ultrasound equipment (Arietta 850, FUJIFILM Healthcare, Tokyo, Japan) equipped with a C252 convex probe (FUJIFILM Healthcare, Tokyo, Japan; frequency range: 1–6 MHz, scan width: 50 mm, scan angle: 70°). Transverse renal ultrasound images were acquired and archived in DICOM format prior to biopsy procedures. All histopathological findings were verified by pathologists with over 10 years of expertise in renal pathology. GN diagnoses were comprehensively categorized based on renal biopsy pathology and authoritative renal pathological classifications [5] and rigorous quality control measures were executed before proceeding with deep learning segmentation and radiomic feature extraction on the renal ultrasound images to eliminate images of inferior quality and those that were blurred or heavily impacted by artifacts.

Renal segmentation

We selected 469 ultrasound images from the dataset for this study. The renal region was manually segmented at the pixel level using the LabelMe tool and the dataset was defined with categorical labels “background” and “kidney”. The labeled image sets were then randomly divided into training and validation sets in an 8:2 ratio. Subsequently, image cropping was standardized to 512 × 512 pixels to ensure consistency of input. During training, a series of preprocessing steps was implemented to enrich data diversity and enhance model generalization. Images were randomly resized while preserving their aspect ratios to incorporate scale variations, followed by stochastic cropping to the preset 512 × 512 pixels. Random flipping and photometric distortions, such as brightness and contrast adjustments, were also applied. The preprocessed images and annotations were then packaged into a standardized input format for model training. A same preprocessing pipeline was adopted for testing. As for the model architecture, an Encoder-Decoder framework incorporating U-Net as the backbone was employed [27]. The U-Net model is a deep learning architecture that has noted advantages in medical image segmentation tasks, able to effectively capture features and detail information and obtaining accurate segmentation in ultrasound and other medical imaging [25]. After iterative training and optimization, the performance of the model was evaluated based on the loss function and the accuracy of the training set, and the best performing model, selected based on its performance indicators on the validation set, especially the key indicator of mean intersection over union (mIoU), was chosen as the final model. Intersection over union (IoU) measures the degree of overlap between the regions predicted by the model and the real labels, reflecting the segmentation accuracy of the model for each category. The mIoU, as the average of the IoUs for each category, serves as an indicator for comprehensive evaluation of model performance [28]. Selecting the model with the highest mIoU value maximizes segmentation accuracy and completeness and ensures generalizability and stability, avoiding overfitting or underfitting [28, 29].

Feature extraction, selection, and modelling

The renal segments of ultrasound images that were segmented from our dataset included127cases of IgAN, 83 cases of MCD, and 103 cases of HSPN. By utilizing the pyradiomics library [30], we extracted 1422 features from each segmented renal portion, incorporating first order features, shape features, gray level co-occurrence matrix (GLCM) features, gray level size zone matrix (GLSZM) features, gray level run length matrix (GLRLM) features, neighbouring gray tone difference matrix (NGTDM) features, and gray level dependence matrix (GLDM) features. These features offer quantitative information about the ultrasound images, including gray level distribution, texture characteristics, shape features, region size and distribution, as well as gray level dependency relationships. This complete set of information is pivotal for disease differentiation, classification, staging, and assessment of treatment response [18,19,20]. Detailed explanations of these specific features are presented in Table 1.

Table 1 Classification and specific description of radiomics features
Full size table

Subsequently, the dataset was randomly partitioned into training and validation sets in an 8:2 split. To identify features that exhibited significant variation across the three GN groups, we employed the analysis of variance (ANOVA) test. Furthermore, we used the supervised least absolute shrinkage and selection operator (LASSO) regression, in conjunction with k-fold cross-validation, to determine the optimal subset of features within one standard error [31]. Finally, a cluster heatmap was generated to visualize these discriminatory radiomics features.

We developed a random forest (RF) classification model with the selected features and thoroughly evaluated its performance by using cross-validation techniques to measure accuracy, sensitivity, and specificity. The model’s parameters and architecture were iteratively optimized, culminating in a final model validated on a separate test dataset. Additionally, the area under the receiver operating characteristic (ROC) curve for each class was calculated and plotted, and a confusion matrix was generated to further assess the model’s predictive capabilities.

Statistical analysis

Python 3.9 and PyTorch 1.12.0 + cu113 were employed for the deep learning model, leveraging the computational power of an NVIDIA GeForce RTX 4090 graphics card. ANOVA was performed to assess the differences among the three groups, with statistical significance set at p < 0.05. Radiomic feature extraction was carried out using the Pyradiomics package. ANOVA, LASSO regression, and RF classification were conducted using the standard functionalities of the Pandas, NumPy, Scikit-learn, and Matplotlib libraries.

Results

Overview of the workflow

We selected 469 renal ultrasound images and divided them into training and validation sets in an 8:2 ratio to train a U-Net model for precise renal ultrasound image segmentation. Leveraging radiomics technology, we extracted numerous radiomic features from the segmented renal regions and categorized the extracted features according to the type of nephropathy (IgAN, 127 cases; MCD, 83 cases; and HSPN, 103 cases). These groups were further randomly split into training and validation sets at an 8:2 ratio. Within the training set, ANOVA was employed for feature selection, followed by LASSO regression for dimensionality reduction, ultimately leading to the selection of 37 features. These features were then integrated with the RF algorithm to develop a classification model for GN. The performance of this model was thoroughly evaluated using the validation set. The complete flowchart is shown in Fig. 1.

Fig. 1
figure 1

Overview of the automatic identification and classification of children's glomerulonephritis ultrasound images. A. Renal Ultrasound Image Segmentation via U-Net; B Classification of Renal Ultrasound Images Based on Pathological Outcomes; C Feature Selection through LASSO Regression; D Modeling with Random Forests; E Evaluation of Training Set Efficacy in Random Forest Modeling; F Assessment of Validation Set Efficacy in Random Forest Methodologies

Full size image

Evaluation of the image segmentation model

Figure 2A depicts the evolution of the loss function in the training set over various steps. The horizontal axis spans from 0 to 40,000 steps, while the vertical axis displays values for evaluation metrics, including loss, ‘decode.loss_ce’, and ‘aux.loss_ce’, represented by red, green, and blue lines, respectively. Each metric exhibits a decreasing trend as the training proceeds, stabilizing at low levels. Figure 2B specifically highlights the training set accuracy trend over multiple iterations, showing the gradual improvement of the accuracy throughout training, albeit with minor fluctuations indicative of the model’s continuous learning and optimization. Figure 2C shows the performance of various evaluation metrics on the test set across steps, encompassing average accuracy (aAcc), mIoU, mean accuracy (mAcc), mean Dice coefficient (mDice), mean F score (mFscore), mean precision (mPrecision), and mean recall (mRecall). A consistent improvement across metrics leads to a relatively stable performance. Table 2 details the optimal model performance, revealing excellent segmentation results for both the “background” and “kidney” categories. Additionally, Fig. 2D presents a randomly selected segmented test image along with a confusion matrix plot, exhibiting excellent performance of various indicators.

Fig. 2
figure 2

Evaluation of the image segmentation model. A Training Set Loss Evolution; B Training Set Performance Metrics; C, Validation Set Performance Metrics; D Randomized Test Sample and Confusion Matrix

Full size image
Table 2 Performance evaluation of the final selected model
Full size table

Feature extraction, selection, and modelling

By using radiomics feature extraction, we identified a total of 1422 features encompassing morphology, first-order statistics, second- and higher-order textures, and wavelets. These features were then analyzed by ANOVA, followed by feature selection using LASSO regression. Lambda was selected as the point where the mean squared error (MSE) is minimized, corresponding to one standard error from the minimum MSE, ultimately leading to the selection of 37 features. As shown in Fig. 3A, the model complexity initially decreased as the regularization parameter Lambda increased in LASSO regression, resulting in a corresponding reduction in the MSE. However, additional increases in Lambda led to an increased MSE, indicating potential overfitting. In Fig. 3B, the vertical dashed line on the right indicates Lambda = 1 × the standard error and serves as a reference for balancing model complexity with predictive performance. Figure 3C presents a cluster heatmap of the 37 selected features, displaying the expression level differences between groups.

Fig. 3
figure 3

LASSO regression and feature cluster heatmap. A the MSE-Lambda Relationship: Lambda at One Standard Error, B Feature Selection Variation with Lambda, C the Cluster Heatmap of the Selected 37 Features

Full size image

We developed a predictive model based on 37 selected features by using the RF algorithm and evaluated the model’s performance through ROC curves and confusion matrices. In the training set, the model demonstrated excellent classification capabilities, with area under the ROC (AUC) values of 0.97 for both HSPN and MCD categories, and 0.93 for IgAN. These high AUC values were indicative of the strong predictive power of the model across various categories. The AUC values remained consistently high in the validation set (HSPN, 0.98; MCD,0.91; IgAN, 0.94), further supporting the robust performance of the model. The visual representation of the model’s classification accuracy provided by the confusion matrix represents high precision and recall in predicting HSPN and IgAN, with slightly lower performance in predicting MCD. Overall, our RF model exhibited outstanding classification prediction capabilities, suggesting significant potential for providing clinicians and researchers with valuable reference information (Fig. 4).

Fig. 4
figure 4

ROC curve and confusion matrix. A ROC curve of the training set; B ROC curve of the validation set; C, confusion matrix of the validation set

Full size image

Discussion

GN is a prevalent kidney disease. The severity varies based on etiology, disease duration, renal dysfunction, and the occurrence of complications [32]. The incidence of GN diagnosis in children has been steadily rising [33]. The early symptoms are often subtle, leading to potential for overlooked diagnosis and missed optimal treatment windows [34]. While renal biopsy will remain the gold standard for GN diagnosis [35], the procedure can be cumbersome, risky, and time-consuming [36], and since early and accurate diagnosis of GN is crucial for effective therapeutic intervention and improved patient outcomes [32], the aim of this study was to construct a renal radiomics model for the early prediction of GN thorough automatically and accurately segmenting and extracting the imaging histological features in order to provide an effective diagnostic adjunct for clinical decision making.

In this study, pediatric pre-renal biopsy renal ultrasound images were collected to form a dataset of renal ultrasound images. A U-Net model was then trained using 469 images to achieve precise segmentation of the renal portions. Employing radiomics methodology, a variety of radiomic features were extracted from these segmented areas. Algorithms including ANOVA, LASSO regression, and RF were leveraged to construct a three-way classification model capable of distinguishing between IgAN, MCD, and HSPN groups. Notably, our approach differs from previous studies that often relied on manual delineation of ROIs and focused on a narrower spectrum of pathological types [11, 37]. Instead, our model automatically delineates ROIs, exhibiting superior intelligence and diagnostic efficacy. The findings suggest that a machine learning-based radiomics model has the potential to accurately predict the pathological types of GN in pediatric patients. This model can facilitate rapid disease assessment, aid in the development of personalized treatment strategies, and ultimately enhance treatment outcomes.

In this study, the U-Net model was employed for semantic segmentation of kidney ultrasound images, leveraging its widespread use and excellent performance in medical imaging, particularly in handling the complex structures and textures inherent in kidney ultrasound images [38]. The U-Net's end-to-end training approach enables automatic learning of optimal feature representations, eliminating the need for manual feature extractor design, thereby suiting our research objectives [39]. Post-training, the model exhibited remarkable kidney region segmentation performance, achieving an accuracy of 95.19% and validating its high precision. Given the excellent performance demonstrated by our preliminary results, we have decided to keep the U-Net model backbone network structure unchanged, with a focus on refining the data preprocessing steps and optimization techniques to further enhance the segmentation performance.

For feature selection, we utilized ANOVA, an initial feature selection technique. ANOVA is a robust statistical tool that excels in detecting statistically significant differences among three or more independent samples [40]. For this study, ANOVA efficiently pinpointed, via F-statistics and p values, radiomic features displaying significant variations across groups. This pivotal step facilitated the preliminary screening of an extensive feature pool (exceeding 1000 features per group), isolating those most likely to contribute meaningfully to classification. By mitigating the “curse of dimensionality” inherent in high-dimensional data, ANOVA-based selection reduced model complexity and computational demands [41]. This streamlined the analysis by eliminating redundant or non-discriminatory features, thereby enhancing model interpretability and predictive accuracy. However, while ANOVA successfully identified features exhibiting statistically significant differences among groups, it could not directly assess the strength of their linear relationships with the response variable. LASSO regression, a sparse linear regression technique, through the incorporation of an L1 regularization term, inherently performs feature selection during model fitting [42]. It systematically shrinks the coefficients of features that are less relevant to the prediction task towards zero, effectively pruning the feature space. This not only leads to a more concise model but also enhances its interpretability by focusing on the most salient predictors [43]. Crucially, the regularization property of LASSO regression mitigates the risk of overfitting in scenarios with a high feature-to-sample ratio. Even after the initial reduction by ANOVA, 800 significant features remained. This number far exceeded the number of observations. By constraining the model complexity, LASSO regression enhanced the generalization capability of the resultant model, ensuring robust performance on unseen data. After initial feature screening via ANOVA, the selection process was further refined by adhering to the one standard error rule and employing LASSO regression, ultimately resulting in the identification of 37 features. This selection was congruent with the structure of the dataset and the intricate interplay between predictor variables and the outcome, thereby enabling the development of an effective yet concise predictive model that harmoniously balanced predictive capability with interpretability [44].

RF is an ensemble learning method that constructs a multitude of decision trees during training and outputs the class that is the mode of the classes or mean prediction of the individual trees. Each tree is built using a random sample of the data, with replacement (bootstrap sampling), and when splitting a node, the best split is chosen from a random subset of the features [45]. RF excels in achieving high accuracy with minimal tuning, is naturally robust against overfitting, offers a built-in feature importance assessment, scales well to large datasets, and is user-friendly, making it a versatile and popular choice for practitioners and researchers alike [46].

It is crucial to acknowledge several potential limitations in our study. Firstly, the relatively limited data sources may have introduced some regional bias to our findings. Furthermore, we focused only on three pathological subtypes of GN, which inevitably limited the scope of the model. To overcome these limitations, we intend to broaden our database by incorporating ultrasound image data from multiple centers and a wider range of cases. Additionally, we plan to explore more methods for image feature extraction and consider various machine learning algorithms. These enhancements will enable us to validate and optimize our model further, ultimately enhancing its accuracy and reliability in real-world applications.

Conclusion

In summary, the study shows that a machine learning model, established using renal ultrasound images, can accurately differentiate between IgAN, MCD, and HSPN in pediatric patients. This can aid in the clinical diagnosis and treatment of GN.

Data availability

No datasets were generated or analysed during the current study.

Abbreviations

GN:

Glomerulonephritis

IgAN:

IgA nephropathy

MCD:

Minimal change disease

HSPN:

Henoch–Schönlein purpura nephritis

MN:

Membranous nephropathy

FSGS:

Focal segmental glomerulosclerosis

MPGN:

Membranoproliferative glomerulonephritis

ROIs:

Regions of interest

mIoU:

Mean intersection over union

IoU:

Intersection over union

ANOVA:

Analysis of variance

LASSO:

The supervised least absolute shrinkage and selection operator

RF:

Random tree

ROC:

Receiver operating characteristic

GLCM:

Gray level co-occurrence matrix

GLSZM:

Gray level size zone matrix

GLRLM:

Gray level run length matrix

NGTDM:

Neighbouring gray tone difference matrix

GLDM:

Gray level dependence matrix

References

  1. Zhan HQ, Zhang X, Chen XL, et al. Application of nanotechnology in the treatment of glomerulonephritis: current status and future perspectives. J Nanobiotechnol. 2024;22(1):9.

    Article  Google Scholar 

  2. Sethi S, Fervenza FC. Standardized classification and reporting of glomerulonephritis. Nephrol Dial Transplant. 2019;34(2):193–9.

    Article  Google Scholar 

  3. Anders HJ, Kitching AR, Leung N, et al. Glomerulonephritis: immunopathogenesis and immunotherapy. Nat Rev Immunol. 2023;23(7):453–71. https://doi.org/10.1038/s41577-022-00816-y.

    Article  Google Scholar 

  4. Glassock RJ. Precision medicine for the treatment of glomerulonephritis: a bold goal but not yet a transformative achievement. Clin Kidney J. 2021;15(4):657–62.

    Article  Google Scholar 

  5. Kidney Disease: Improving Global Outcomes (KDIGO) Glomerular Diseases Work Group. KDIGO 2021 Clinical Practice Guideline for the Management of Glomerular Diseases. Kidney Int. 2021;100(4S):S1–276.

    Google Scholar 

  6. Sethi S, De Vriese AS, Fervenza FC. Acute glomerulonephritis. Lancet. 2022;399(10335):1646–63.

    Article  Google Scholar 

  7. Serhal M, Rangwani S, Seedial SM, et al. Safety and diagnostic efficacy of image-guided biopsy of small renal masses. Cancers. 2024;16(4):835.

    Article  Google Scholar 

  8. Patel AG, Kriegshauser JS, Young SW, et al. Detection of bleeding complications after renal transplant biopsy. AJR Am J Roentgenol. 2021;216(2):428–35.

    Article  Google Scholar 

  9. Zhan T, Lou A. Comparison of outcomes of an 18-gauge vs 16-gauge ultrasound-guided percutaneous renal biopsy: a systematic review and meta-analysis. Ren Fail. 2023;45(2):2257806.

    Article  Google Scholar 

  10. Qin X, Xia L, Ma Q, et al. Development of a novel combined nomogram model integrating deep learning radiomics to diagnose IgA nephropathy clinically. Ren Fail. 2023;45(2):2271104.

    Article  Google Scholar 

  11. Zhang L, Chen Z, Feng L, et al. Preliminary study on the application of renal ultrasonography radiomics in the classification of glomerulopathy. BMC Med Imaging. 2021;21(1):115. https://doi.org/10.1186/s12880-021-00647-8.

    Article  Google Scholar 

  12. Shang J, Zhang Y, Guo R, et al. Gut microbiome analysis can be used as a noninvasive diagnostic tool and plays an essential role in the onset of membranous nephropathy. Adv Sci. 2022;9(28):e2201581.

    Article  Google Scholar 

  13. Yang S, Gill PJ, Anwar MR, et al. Kidney ultrasonography after first febrile urinary tract infection in children: a systematic review and meta-analysis. JAMA Pediatr. 2023;177(8):764–73.

    Article  Google Scholar 

  14. Bodard S, Denis L, Hingot V, et al. Ultrasound localization microscopy of the human kidney allograft on a clinical ultrasound scanner. Kidney Int. 2023;103(5):930–5.

    Article  Google Scholar 

  15. Nicolau C, Antunes N, Paño B, et al. Imaging characterization of renal masses. Medicina (Kaunas). 2021;57(1):51.

    Article  Google Scholar 

  16. Caroli A, Remuzzi A, Lerman LO. Basic principles and new advances in kidney imaging. Kidney Int. 2021;100(5):1001–11.

    Article  Google Scholar 

  17. Chen M, Copley SJ, Viola P, et al. Radiomics and artificial intelligence for precision medicine in lung cancer treatment. Semin Cancer Biol. 2023;93:97–113.

    Article  Google Scholar 

  18. Feng Z, Li H, Liu Q, et al. CT Radiomics to predict macrotrabecular-massive subtype and immune status in hepatocellular carcinoma. Radiology. 2023;307(1):e221291.

    Article  Google Scholar 

  19. Wang T, She Y, Yang Y, et al. Radiomics for survival risk stratification of clinical and pathologic stage IA pure-solid non-small cell lung cancer. Radiology. 2022;302(2):425–34.

    Article  Google Scholar 

  20. Crispin-Ortuzar M, Woitek R, Reinius MAV, et al. Integrated radiogenomics models predict response to neoadjuvant chemotherapy in high grade serous ovarian cancer. Nat Commun. 2023;14(1):6756.

    Article  Google Scholar 

  21. Yu Y, Tan Y, Xie C, et al. Development and validation of a preoperative magnetic resonance imaging radiomics-based signature to predict axillary lymph node metastasis and disease-free survival in patients with early-stage breast cancer. JAMA Netw Open. 2020;3(12):e2028086.

    Article  Google Scholar 

  22. Morales MA, Manning WJ, Nezafat R. Present and future innovations in AI and cardiac MRI. Radiology. 2024;310(1):e231269.

    Article  Google Scholar 

  23. Zhang R, He Z, Zhu J, et al. Collaborative multi-feature extraction and scale-aware semantic information mining for medical image segmentation. Phys Med Biol. 2022;67(20):205008. https://doi.org/10.1088/1361-6560/ac95f5.

    Article  Google Scholar 

  24. Zhou S, Nie D, Adeli E, et al. Semantic instance segmentation with discriminative deep supervision for medical images. Med Image Anal. 2022;82:102626.

    Article  Google Scholar 

  25. Chen G, Li L, Dai Y, et al. AAU-Net: an adaptive attention U-Net for breast lesions segmentation in ultrasound images. IEEE Trans Med Imaging. 2023;42(5):1289–300.

    Article  Google Scholar 

  26. Xun S, Li D, Zhu H, et al. Generative adversarial networks in medical image segmentation: a review. Comput Biol Med. 2022;140:105063.

    Article  Google Scholar 

  27. Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation. In: International Conference on Medical image computing and computer-assisted intervention. Springer, Cham; 2015. p. 234–241.

  28. Madani A, Namazi B, Altieri MS, et al. Artificial intelligence for intraoperative guidance: using semantic segmentation to identify surgical anatomy during laparoscopic cholecystectomy. Ann Surg. 2022;276(2):363–9.

    Article  Google Scholar 

  29. Shi C, Wang K, Zhang G, et al. Efficient and accurate semi-supervised semantic segmentation for industrial surface defects. Sci Rep. 2024;14(1):21874.

    Article  Google Scholar 

  30. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 2017;77(21):e104–7.

    Article  Google Scholar 

  31. Zheng YM, Chen J, Xu Q, et al. Development and validation of an MRI-based radiomics nomogram for distinguishing Warthin’s tumour from pleomorphic adenomas of the parotid gland. Dentomaxillofac Radiol. 2021;50(7):20210023.

    Article  Google Scholar 

  32. Kronbichler A, Bajema I, Geetha D, et al. Novel aspects in the pathophysiology and diagnosis of glomerular diseases. Ann Rheum Dis. 2023;82(5):585–93.

    Article  Google Scholar 

  33. Boussetta A, Ben Njima K, Jellouli M, et al. Primary focal segmental glomerulosclerosis in children: epidemiology, clinical presentation and prognosis. Tunis Med. 2023;101(3):373–8.

    Google Scholar 

  34. Quinn GZ, Abedini A, Liu H, et al. Renal histologic analysis provides complementary information to kidney function measurement for patients with early diabetic or hypertensive disease. J Am Soc Nephrol. 2021;32(11):2863–76.

    Article  Google Scholar 

  35. Zee J, Liu Q, Smith AR, et al. Kidney biopsy features most predictive of clinical outcomes in the spectrum of minimal change disease and focal segmental glomerulosclerosis. J Am Soc Nephrol. 2022;33(7):1411–26.

    Article  Google Scholar 

  36. Ho QY, Lim CC, Tan HZ, et al. Complications of percutaneous kidney allograft biopsy: systematic review and meta-analysis. Transplantation. 2022;106(7):1497–506.

    Article  Google Scholar 

  37. Chen J, Wen Z, Yang X, et al. Ultrasound-based radiomics for the classification of Henoch–Schönlein purpura nephritis in children. Ultrason Imaging. 2024;46(2):110–20.

    Article  Google Scholar 

  38. Valente S, Morais P, Torres HR, et al. A deep learning method for kidney segmentation in 2D ultrasound images. In: 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC),vol. 2022; 2022. p. 3911–3914

  39. Wu H, Souedet N, Jan C, et al. A general deep learning framework for neuron instance segmentation based on efficient UNet and morphological post-processing. Comput Biol Med. 2022;150:106180.

    Article  Google Scholar 

  40. Bommert A, Welchowski T, Schmid M, et al. Benchmark of filter methods for feature selection in high-dimensional gene expression survival data. Brief Bioinform. 2022;23(1):bbab354.

    Article  Google Scholar 

  41. Laghari AA, Estrela V, Yin S. How to collect and interpret medical pictures captured in highly challenging environments that range from nanoscale to hyperspectral imaging. Curr Med Imaging. 2024;20:e281222212228.

    Article  Google Scholar 

  42. Narala S, Li SQ, Klimas NK, et al. Application of least absolute shrinkage and selection operator logistic regression for the histopathological comparison of chondrodermatitis nodularis helicis and hyperplastic actinic keratosis. J Cutan Pathol. 2021;48(6):739–44.

    Article  Google Scholar 

  43. Ajith S, Debnath MK, Karthik R. Statistical and machine learning models for location-specific crop yield prediction using weather indices. Int J Biometeorol. Published online August 31, 2024.

  44. Wind M, Fierro JJ, Bloemenkamp KWM, et al. Pregnancy outcome predictors in systemic lupus erythematosus: a systematic review and meta-analysis. Lancet Rheumatol. 2024;6(10):e667–83.

    Article  Google Scholar 

  45. Steinbach D, Ahrens PC, Schmidt M, et al. Applying machine learning to blood count data predicts sepsis with ICU admission. Clin Chem. 2024;70(8):1088.

    Article  Google Scholar 

  46. Villani A, Fontana A, Panebianco C, et al. A powerful machine learning approach to identify interactions of differentially abundant gut microbial subsets in patients with metastatic and non-metastatic pancreatic cancer. Gut Microbes. 2024;16(1):2375483.

    Article  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

This research received no specific grants from any funding agencies in the public, commercial, or not-for-profit sectors.

Author information

Author notes

  1. Jun Kou, Zuying Li and Yazi You have contributed equally to this work.

Authors and Affiliations

  1. Department of Ultrasound, Children’s Hospital of Chongqing Medical University, National Clinical Research Center for Child Health and Disorders, Ministry of Education Key Laboratory of Child Development and Disorders, Chongqing Key Laboratory of Pediatric Metabolism and Inflammatory Diseases, Chongqing, 400010, China

    Jun Kou, Zuying Li, Yazi You, Ruiqi Wang, Jingyu Chen & Yi Tang

Authors
  1. Jun Kou
    View author publications

    Search author on:PubMed Google Scholar

  2. Zuying Li
    View author publications

    Search author on:PubMed Google Scholar

  3. Yazi You
    View author publications

    Search author on:PubMed Google Scholar

  4. Ruiqi Wang
    View author publications

    Search author on:PubMed Google Scholar

  5. Jingyu Chen
    View author publications

    Search author on:PubMed Google Scholar

  6. Yi Tang
    View author publications

    Search author on:PubMed Google Scholar

Contributions

Jou Kou, Zuying Li and Yazi You: Conception and design of study, acquisition of data, analysis and/or interpretation of data, drafting the manuscript, approval of the version of the manuscript to be published. Jingyu Chen and Ruiqi Wang: acquisition of data, revising the manuscript critically for important intellectual content, approval of the version of the manuscript to be published. Yi Tang: Conception and design of study, acquisition of data, revising the manuscript critically for important intellectual content, approval of the version of the manuscript to be published.

Corresponding author

Correspondence to Yi Tang.

Ethics declarations

Ethics approval and consent to participate

Before participation, all involved individuals carefully read and signed an informed consent form approved by the Ethics Committee. At the same time, the Ethics Committee of the hospital (Children’s Hospital of Chongqing Medical University) approved the clinical study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Kou, J., Li, Z., You, Y. et al. Automatic identification and classification of pediatric glomerulonephritis on ultrasound images based on deep learning and radiomics. J Big Data 11, 165 (2024). https://doi.org/10.1186/s40537-024-01033-1

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40537-024-01033-1

Keywords