MAL-Net: A Multi-Label Deep Learning Framework Integrating LSTM and Multi-Head Attention for Enhanced Classification of IgA Nephropathy Subtypes Using Clinical Sensor Data
Abstract
:1. Introduction
2. Materials and Methods
2.1. Patient Selection and Data Collection
2.2. Subtype Classification
2.3. Predictor Variables
2.4. Data Preprocessing
2.4.1. Missing Data Handling
2.4.2. Feature Encoding
2.4.3. Feature Normalization
2.4.4. Class Imbalance Mitigation
2.4.5. Training and Test Sets
2.5. Model Development: MAL-Net Architecture
2.5.1. Memory Network Module (LSTM Layer)
2.5.2. MHA
2.5.3. Output Layer
2.5.4. Loss Function
3. Results
3.1. Experimental Setting
3.1.1. Implementation Details
3.1.2. Evaluation Metrics
3.2. Ablation Study
3.2.1. Performance of MHA in Single-Label Prediction
3.2.2. Hyperparameter Optimization
3.3. Comparative Evaluation of Classification Models
3.4. Performance Evaluation of Multi-Label Classification
4. Discussion
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Schena, F.P.; Nistor, I. Epidemiology of IgA nephropathy: A global perspective. In Seminars in Nephrology; WB Saunders: Philadelphia, PA, USA, 2018; Volume 38, pp. 435–442. [Google Scholar]
- Rajasekaran, A.; Julian, B.A.; Rizk, D.V. IgA nephropathy: An interesting autoimmune kidney disease. Am. J. Med. Sci. 2021, 361, 176–194. [Google Scholar] [CrossRef] [PubMed]
- Magistroni, R.; D’Agati, V.D.; Appel, G.B. New developments in the genetics, pathogenesis, and therapy of IgA nephropathy. Kidney Int. 2015, 88, 974–989. [Google Scholar] [PubMed]
- Diciolla, M.; Binetti, G.; Di Noia, T. Patient classification and outcome prediction in IgA nephropathy. Comput. Biol. Med. 2015, 66, 278–286. [Google Scholar] [PubMed]
- Ying, Y.; Wang, L.; Ma, S. An enhanced machine learning approach for effective prediction of IgA nephropathy patients with severe proteinuria based on clinical data. Comput. Biol. Med. 2024, 173, 108341. [Google Scholar] [CrossRef]
- Rodrigues, J.C.; Haas, M.; Reich, H.N. IgA nephropathy. Clin. J. Am. Soc. Nephrol. 2017, 12, 677–686. [Google Scholar] [CrossRef]
- Trimarchi, H.; Barratt, J.; Cattran, D.C. Oxford classification of IgA nephropathy 2016: An update from the IgA nephropathy classification working group. Kidney Int. 2017, 91, 1014–1021. [Google Scholar] [CrossRef]
- Konieczny, A.; Stojanowski, J.; Krajewska, M. Machine learning in prediction of IgA nephropathy outcome: A comparative approach. J. Pers. Med. 2021, 11, 312. [Google Scholar] [CrossRef]
- Ravindra, B.V.; Sriraam, N.; Geetha, M. Classification of non-chronic and chronic kidney disease using SVM neural networks. Int. J. Eng. Technol. 2018, 7, 191–194. [Google Scholar]
- Ren, Y.; Fei, H.; Liang, X. A hybrid neural network model for predicting kidney disease in hypertension patients based on electronic health records. BMC Med. Inform. Decis. Mak. 2019, 19, 131–138. [Google Scholar] [CrossRef]
- Almasoud, M.; Ward, T.E. Detection of chronic kidney disease using machine learning algorithms with least number of predictors. Int. J. Soft Comput. Its Appl. 2019, 10, 89–96. [Google Scholar] [CrossRef]
- Geng, Y.H.; Zhang, Z.; Zhang, J.J. Established the first clinical prediction model regarding the risk of hyperuricemia in adult IgA nephropathy. Int. Urol. Nephrol. 2023, 55, 1787–1797. [Google Scholar] [CrossRef] [PubMed]
- Li, F.; Wei, R.; Wang, Y. Nomogram prediction model for renal anaemia in IgA nephropathy patients. Open Med. 2021, 16, 718–727. [Google Scholar] [CrossRef] [PubMed]
- Zhuang, K.; Wang, W.; Xu, C. Machine learning-based diagnosis and prognosis of IgAN: A systematic review and meta-analysis. Heliyon 2024, 10, e33090. [Google Scholar] [CrossRef] [PubMed]
- Pattrapornpisut, P.; Avila-Casado, C.; Reich, H.N. IgA nephropathy: Core curriculum 2021. Am. J. Kidney Dis. 2021, 78, 429–441. [Google Scholar] [CrossRef]
- Ren, H.; Lv, W.; Shang, Z. Identifying functional subtypes of IgA nephropathy based on three machine learning algorithms and WGCNA. BMC Med. Genom. 2024, 17, 61. [Google Scholar] [CrossRef]
- Barbour, S.J.; Espino-Hernandez, G.; Reich, H.N. The MEST score provides earlier risk prediction in IgA nephropathy. Kidney Int. 2016, 89, 167–175. [Google Scholar] [CrossRef]
- Schena, F.P.; Anelli, V.W.; Trotta, J. Development and testing of an artificial intelligence tool for predicting end-stage kidney disease in patients with immunoglobulin A nephropathy. Kidney Int. 2021, 99, 1179–1188. [Google Scholar] [CrossRef]
- Venkatesan, V.K.; Ramakrishna, M.T.; Izonin, I. Efficient data preprocessing with ensemble machine learning technique for the early detection of chronic kidney disease. Appl. Sci. 2023, 13, 2885. [Google Scholar] [CrossRef]
- Affonso, C.; Rossi, A.L.D.; Vieira, F.H.A. Deep learning for biological image classification. Expert Syst. Appl. 2017, 85, 114–122. [Google Scholar] [CrossRef]
- Wang, W.F.; Qiu, X.H.; Chen, C.S. Application research on long short-term memory network in fault diagnosis. In Proceedings of the 2018 International Conference on Machine Learning and Cybernetics (ICMLC), Chengdu, China, 15–18 July 2018; Volume 2, pp. 360–365. [Google Scholar]
- Men, L.; Ilk, N.; Tang, X. Multi-disease prediction using LSTM recurrent neural networks. Expert Syst. Appl. 2021, 177, 114905. [Google Scholar] [CrossRef]
- Yang, F.; Zhang, J.; Chen, W. DeepMPM: A mortality risk prediction model using longitudinal EHR data. BMC Bioinform. 2022, 23, 423. [Google Scholar] [CrossRef] [PubMed]
- Khan, M.A.; Kim, Y. Cardiac arrhythmia disease classification using LSTM deep learning approach. Comput. Mater. Contin. 2021, 67, 427. [Google Scholar]
- Sudha, V.K.; Kumar, D. Hybrid CNN and LSTM network for heart disease prediction. SN Comput. Sci. 2023, 4, 172. [Google Scholar] [CrossRef]
- Balaji, E.; Brindha, D.; Elumalai, V.K. Automatic and non-invasive Parkinson’s disease diagnosis and severity rating using LSTM network. Appl. Soft Comput. 2021, 108, 107463. [Google Scholar]
- Choi, E.; Bahadori, M.T.; Sun, J. Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. Adv. Neural Inf. Process. Syst. 2016, 29, 3512–3520. [Google Scholar]
- Vaswani, A. Attention is all you need. Adv. Neural Inf. Process. Syst. 2017, 30, 5998–6008. [Google Scholar]
- Kingma, D.P.; Ba, J. Adam: A method for stochastic optimization. arXiv 2014, arXiv:1412.6980. [Google Scholar]
- Ghojogh, B.; Ghodsi, A. Recurrent neural networks and long short-term memory networks: Tutorial and survey. arXiv 2023, arXiv:2304.11461. [Google Scholar]
- Srivastava, N.; Hinton, G.; Krizhevsky, A. Dropout: A simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 2014, 15, 1929–1958. [Google Scholar]
- Du, J.; Chen, Q.; Peng, Y. ML-Net: Multi-label classification of biomedical texts with deep neural networks. J. Am. Med. Inform. Assoc. 2019, 26, 1279–1285. [Google Scholar] [CrossRef]
- Guo, X.; Chen, L.; Shen, C. Hierarchical adaptive deep convolution neural network and its application to bearing fault diagnosis. Measurement 2016, 93, 490–502. [Google Scholar] [CrossRef]
- Noda, R.; Ichikawa, D.; Shibagaki, Y. Machine learning-based diagnostic prediction of IgA nephropathy: Model development and validation study. Sci. Rep. 2024, 14, 12426. [Google Scholar] [CrossRef] [PubMed]
Characteristic | Training Set (n = 400) | Test Set | p-Value |
---|---|---|---|
(n = 100) | |||
Age | 38.88 ± 0.602 | 40.84 ± 1.067 | 0.084 |
Sex | Male (42.40%) Female (57.6%) | Male (44.80%) Female (55.20%) | 0.640 |
SBP | 125.15 ± 1.018 | 125.70 ± 1.639 | 0.577 |
DBP | 77.52 ± 0.676 | 78.04 ± 1.241 | 0.629 |
UPC | 1.53 ± 0.088 | 1.67 ± 0.225 | 0.973 |
DRBCs | 0 (28.90%) 1 (12.20%) 2 (19.30%) 3 (32.80%) 4 (6.80%) | 0 (28.70%) 1 (13.00%) 2 (18.30%) 3 (32.20%) 4 (7.80%) | 0.993 |
eGFR | 86.54 ± 1.638 | 83.36 ± 3.215 | 0.289 |
Lumbago | 0 (63.30%) 1 (36.70%) | 0 (59.50%) 1 (40.50%) | 0.456 |
Fatigue/tiredness | 0 (44.40%) 1 (55.60%) | 0 (35.30%) 1 (64.70%) | 0.082 |
Susceptible to colds | 0 (76.70%) 1 (23.30%) | 0 (81.00%) 1 (19.00%) | 0.331 |
edema | 0 (60.30%) 1 (39.70%) | 0 (66.00%) 1 (44.00%) | 0.456 |
Halitosis | 0 (56.80%) 1 (43.20%) | 0 (59.50%) 1 (40.50%) | 0.615 |
Spontaneous/night sweats | 0 (78.60%) 1 (21.40%) | 0 (82.80%) 1 (17.20%) | 0.325 |
Infection Risk | 0 (66.10%) 1 (33.90%) | 0 (68.10%) 1 (31.90%) | 0.696 |
Skin rash | 0 (63.00%) 1 (37.00%) | 0 (59.50%) 1 (40.50%) | 0.487 |
Muscle/body/joint soreness | 0 (63.80%) 1 (36.20%) | 0 (63.80%) 1 (36.20%) | 0.995 |
Shortness of breath | 0 (67.40%) 1 (32.60%) | 0 (74.10%) 1 (25.90%) | 0.171 |
Nocturnal urine output ≥ 750 mL) | 0 (65.40%) 1 (34.60%) | 0 (67.20%) 1 (32.80%) | 0.710 |
Heat in palms and soles | 0 (64.60%) 1 (35.40%) | 0 (67.20%) 1 (32.80%) | 0.600 |
Dry eyes | 0 (65.10%) 1 (34.90%) | 0 (67.20%) 1 32.80%) | 0.673 |
Dry throat | 0 (57.10%) 1 (42.90%) | 0 (58.60%) 1 (41.40%) | 0.772 |
Fixed lower back pain | 0 (54.80%) 1 (45.20%) | 0 (55.20%) 1 (44.80%) | 0.941 |
duration ≥ 3 M | 0 (64.60%) 1 (35.40%) | 0 (65.50%) 1 (34.50%) | 0.856 |
Skin purpura, petechiae/spider veins | 0 (59.70%) 1 (40.30%) | 0 (57.80%) 1 (42.20%) | 0.710 |
Limb numbness | 0 (62.80%) 1 (37.20%) | 0 (57.80%) 1 (42.20%) | 0.328 |
Dusky facial complexion | 0 (55.00%) 1 (45.00%) | 0 (57.80%) 1 (42.20%) | 0.605 |
Irritability and anger | 0 (82.20%) 1 (17.80%) | 0 (83.60%) 1 (16.40%) | 0.089 |
Headache | 0 (80.60%) 1 (19.40%) | 0 (82.80%) 1 (17.20%) | 0.606 |
Blurry or darkened vision | 0 (82.70%) 1 (17.30%) | 0 (87.10%) 1 (12.90%) | 0.262 |
Tremors, cramps | 0 (78.30%) 1 (21.70%) | 0 (81.00%) 1 (19.00%) | 0.099 |
Anorexia, nausea | 0 (96.60%) 1 (3.40%) | 0 (94.00%) 1 (6.00%) | 0.629 |
Dull complexion | 0 (97.40%) 1 (2.60%) | 0 (96.60%) 1 (3.40%) | 0.620 |
Fear of cold | 0 (96.40%) 1 (3.60%) | 0 (94.00%) 1 (6.00%) | 0.254 |
Subtype | Method | Accuracy | Precision | Recall | AUC |
---|---|---|---|---|---|
Qi-Yin Deficiency | Baseline | 0.85 | 0.76 | 0.99 | 0.98 |
MAL-Net | 0.98 | 0.99 | 0.95 | 1.00 | |
Wind-dampness | Baseline | 0.89 | 0.97 | 0.82 | 0.96 |
MAL-Net | 0.92 | 0.97 | 0.89 | 0.99 | |
Blood Stasis | Baseline | 0.83 | 0.80 | 0.99 | 0.80 |
MAL-Net | 0.93 | 0.78 | 0.96 | 0.89 | |
Liver-wind | Baseline | 0.90 | 0.85 | 0.68 | 0.94 |
MAL-Net | 0.86 | 0.93 | 0.68 | 0.94 | |
Ni-du | Baseline | 0.92 | 0.14 | 0.00 | 0.88 |
MAL-Net | 0.93 | 0.98 | 0.67 | 0.96 |
Hyperparameter | Learning Rate | Dropout Rate | Bitch Size | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.001 | 0.0001 | 0.1 | 0.2 | 0.4 | 0.5 | 16 | 32 | 64 | 128 | |
Accuracy (%) | 81.2 | 86.8 | 91.9 | 88.3 | 90.6 | 90.7 | 89.3 | 80.5 | 85.6 | 91.6 | 89.2 |
Recall (%) | 80.5 | 85.7 | 91.2 | 87.4 | 89.7 | 89.8 | 88.5 | 79.6 | 84.5 | 91.0 | 88.3 |
Precision (%) | 82.1 | 87.2 | 92.0 | 88.9 | 91.2 | 91.3 | 89.4 | 81.2 | 86.0 | 92.0 | 89.7 |
F1-score (%) | 81.3 | 86.4 | 91.6 | 88.1 | 90.4 | 90.5 | 88.9 | 80.4 | 85.2 | 91.5 | 89.0 |
Classification Model | Accuracy | Precision | Recall | F1-Score | AUC | p-Value |
---|---|---|---|---|---|---|
RF | 0.892 ± 0.012 | 0.875 ± 0.015 | 0.850 ± 0.018 | 0.862 ± 0.014 | 0.915 ± 0.010 | 0.015 |
SVM | 0.630 ± 0.020 | 0.635 ± 0.022 | 0.645 ± 0.025 | 0.620 ± 0.018 | 0.800 ± 0.015 | 0.036 |
LR | 0.850 ± 0.015 | 0.780 ± 0.020 | 0.760 ± 0.022 | 0.770 ± 0.019 | 0.880 ± 0.012 | 0.029 |
DNN | 0.620 ± 0.025 | 0.320 ± 0.030 | 0.450 ± 0.035 | 0.380 ± 0.028 | 0.780 ± 0.020 | <0.01 |
CNN | 0.670 ± 0.030 | 0.350 ± 0.035 | 0.430 ± 0.040 | 0.390 ± 0.032 | 0.575 ± 0.025 | <0.001 |
LSTM | 0.870 ± 0.011 | 0.700 ± 0.012 | 0.675 ± 0.015 | 0.690 ± 0.013 | 0.920 ± 0.008 | 0.043 |
Proposed Model | 0.905 ± 0.010 | 0.910 ± 0.010 | 0.855 ± 0.012 | 0.885 ± 0.011 | 0.972 ± 0.006 |
Model | Parameters | FLOPs |
---|---|---|
RF | N/A | 50 M |
SVM | N/A | 5 M |
LR | 0.03 K | 0.05 M |
CNN | 15 M | 1.9 G |
DNN | 2.5 M | 500 M |
LSTM (Baseline) | 8.3 K | 0.16 M |
Our Model | 14.6 K | 0.49 M |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Wang, H.; Liao, Y.; Gao, L.; Li, P.; Huang, J.; Xu, P.; Fu, B.; Zhu, Q.; Lai, X. MAL-Net: A Multi-Label Deep Learning Framework Integrating LSTM and Multi-Head Attention for Enhanced Classification of IgA Nephropathy Subtypes Using Clinical Sensor Data. Sensors 2025, 25, 1916. https://doi.org/10.3390/s25061916
Wang H, Liao Y, Gao L, Li P, Huang J, Xu P, Fu B, Zhu Q, Lai X. MAL-Net: A Multi-Label Deep Learning Framework Integrating LSTM and Multi-Head Attention for Enhanced Classification of IgA Nephropathy Subtypes Using Clinical Sensor Data. Sensors. 2025; 25(6):1916. https://doi.org/10.3390/s25061916
Chicago/Turabian StyleWang, Hongyan, Yuehui Liao, Li Gao, Panfei Li, Junwei Huang, Peng Xu, Bin Fu, Qin Zhu, and Xiaobo Lai. 2025. "MAL-Net: A Multi-Label Deep Learning Framework Integrating LSTM and Multi-Head Attention for Enhanced Classification of IgA Nephropathy Subtypes Using Clinical Sensor Data" Sensors 25, no. 6: 1916. https://doi.org/10.3390/s25061916
APA StyleWang, H., Liao, Y., Gao, L., Li, P., Huang, J., Xu, P., Fu, B., Zhu, Q., & Lai, X. (2025). MAL-Net: A Multi-Label Deep Learning Framework Integrating LSTM and Multi-Head Attention for Enhanced Classification of IgA Nephropathy Subtypes Using Clinical Sensor Data. Sensors, 25(6), 1916. https://doi.org/10.3390/s25061916