04055384

October 04, 2025, 15:34

Reader’s expertise on the topic of the manuscript

Does the reader have a conflict of interest?

The overall quality of the manuscript, based on the above-listed criteria, should be evaluated and classified according to the following five categories

Language quality (style, grammar, and spelling) should be evaluated and classified according to the following five categories.

Machine Learning in Surgical Decision-Making and Non-Surgical Treatment of Alveolar Echinococcosis Fengying liu,Yongfang Xie Introduce Alveolar Echinococcosis (AE) is a chronic, progressive liver disease caused by the ingestion of Echinococcus multilocularis larvae, with foxes and wild dogs as definitive hosts and rodents as intermediate hosts1. It typically centers in the liver, slowly spreading and metastasizing, with an incubation period lasting 5 to 15 years. Due to early non-specific symptoms such as fever and weight loss, patients are often difficult to diagnose in the early stages. In the later stages, continuous asexual reproduction of Echinococcus multilocularis and severe inflammatory granulomatous infiltration around the parasite lead to widespread fibrosis and necrosis. Without timely treatment, the mortality rate of patients can reach 90% within 10 years1. Currently, radical surgery combined with adjuvant chemotherapy is the only curative treatment method2. However, most patients are diagnosed at advanced stages, missing the optimal timing for liver resection. Ex vivo liver resection and liver autotransplantation (ELRA)3 offer hope for these patients, but traditional surgical decision-making often relies on the clinical experience of surgeons, which can be subjective and prone to bias. In recent years, machine learning has made significant progress in the medical field, especially in disease prediction, treatment selection, and clinical diagnosis, showing tremendous potential for development. By processing large amounts of clinical data, machine learning uncovers complex underlying relationships and assists doctors in making more accurate and objective decisions. Against this backdrop, I carefully reviewed the study published by Da-Long Zhu et al., and found that in the treatment of Alveolar Echinococcosis (HAE), machine learning not only aids in surgical decision-making but also provides valuable support for patients undergoing non-surgical treatments, driving the development of precision medicine4. 1.The Advantages and Challenges of Machine Learning in the Explainability of Surgical Decision-Making for HAE 1.1 Innovative Analysis of Machine Learning Da-Long Zhu et al. conducted a retrospective cohort study based on data from the First Affiliated Hospital of Xinjiang Medical University, including 710 patients, of whom 545 underwent liver resection and 165 underwent ex vivo liver resection and ELRA. The authors innovatively applied three feature selection techniques—recursive feature elimination, minimum redundancy maximum relevance, and LASSO regression—with cross-validation to enhance the model. Ultimately, they identified 10 key features from the extensive clinical data, including lesion size, vascular invasion type, bile duct invasion degree, white blood cell count, platelet count, hemoglobin, indirect bilirubin, serum creatinine, γ-glutamyl transferase, and prothrombin time. These features are closely related to the prognosis of HAE and provide important data support for surgical decision-making. Da-Long Zhu et al. systematically compared the performance and explainability of 11 machine learning models, ultimately selecting the XGBoost model and using Bayesian optimization for hyperparameter tuning.XGBoost5,as an efficient ensemble learning algorithm, demonstrates strong predictive ability in various classification and regression tasks. Model evaluation showed an AUC of 0.935 for the training set and 0.734 for the validation set, with a sensitivity of 93.6% and a specificity of 90.9%. Both the calibration curve and decision curve displayed good clinical utility, effectively predicting whether a patient is suitable for liver resection or ELRA. Furthermore, XGBoost, with its ensemble learning characteristics, reduces the risk of overfitting, providing a stable decision-making tool. To improve the interpretability of the model, Da-Long Zhu et al. employed the SHAP analysis method5 to explain the contribution of each feature to the prediction. Further analysis of the results revealed that vascular invasion type is a key factor in choosing ELRA, while indicators such as platelet count, prothrombin time, and hemoglobin reflect the patient's coagulation function, liver reserve, and inflammatory status. For instance, when significant invasion of the hepatic vein and inferior vena cava is present, the choice of ELRA is more likely, whereas for patients with no vascular invasion or only portal vein involvement, liver resection is preferred. Through SHAP analysis, the model's prediction process became more transparent, enhancing clinicians' trust in the model. 1.2 Limitations and Challenges of the Model It is commendable that Da-Long Zhu et al. accurately identified the clinical decision-making challenges in the surgical treatment of AE and achieved satisfactory results. However, the model still has some limitations. First, the sample size is relatively small, and the data comes from a single source, which may not fully represent patients from different regions or healthcare settings. Secondly, imaging examinations, such as CT scans, are an important tool for diagnosing HAE. However, the study did not conduct a deeper analysis and evaluation of the quantitative features extracted from imaging omics data, such as tumor morphology, texture, and other imaging features6. According to the research by Yener Aydin et al7, HAE often presents with characteristics such as pulmonary nodules, lesions, cavities, and metastatic tumors in imaging. In clinical practice, small solid nodules are easily confused with malignant tumors. Machine learning can help identify the differences and make accurate judgments. In addition to the XGBoost model used by Da-Long Zhu et al., other machine learning models also have their advantages in clinical prediction tasks. To facilitate the rational selection and comparison of models in this field, the following summary is provided (Table 1). Table 1 Model Advantages Limitations XGBoost8 Robust performance, accurate predictions, and comprehensive functionality, suitable for medium-sized data Slower training speed; risk of overfitting, requires regularization control LightGBM9 Extremely fast training speed, high memory efficiency, suitable for large-scale data Higher risk of overfitting on small datasets CatBoost10 Excellent handling of categorical features, high prediction accuracy, fast training speed, strong anti-overfitting ability Lower model flexibility TabNet11 Attention-based deep learning framework, high performance potential, built-in interpretability Requires large datasets and computational power Therefore, future research should focus on multi-center, large-sample validation to improve the external validity and generalizability of the model. Additionally, extracting quantitative features from imaging omics should be prioritized to enhance the clinical relevance and application value of the model. 2. The Prospects of Machine Learning in Non-Surgical Treatment of HAE 2.1 Current Non-Surgical Treatment Methods for HAE The non-surgical treatment of HAE relies on albendazole-based medications12, which interfere with microtubule formation by binding to β-tubulin, thereby impairing nutrient absorption and parasite growth. However, albendazole does not kill the parasite and has significant hepatotoxicity. Therefore, although the medication can improve patient survival rates, its effectiveness remains limited, especially for patients who cannot undergo surgery. In such cases, prevention and continuous monitoring become crucial treatment strategies. In addition to conventional drug therapy, the development of new drugs is also an important direction for non-surgical treatment. Mefloquine12 has shown certain advantages in terms of its anti-parasitic effect and lower hepatotoxicity.Furthermore, progress has been made in vaccine development, such as the recombinant Tetraspanin 3 vaccine13 , which induces strong local and systemic immune responses, effectively protecting humans from infections of Echinococcus multilocularis in the intestine, bloodstream, and liver. However, these studies have not yet undergone large-scale clinical trials and face challenges regarding individual variations in efficacy. 2.2 The Prospects of Machine Learning in HAE Non-Surgical Treatment Through Imaging and Multi-Omics Data Machine learning also shows great potential in the non-surgical treatment of HAE. By deeply analyzing patients' imaging features and multi-omics data, machine learning can provide precise analysis in early disease diagnosis, treatment processes, and prognosis assessment. Since early symptoms of HAE are often atypical, making early detection difficult, machine learning can significantly enhance the accuracy of early intervention, treatment effectiveness, and prognosis prediction through the deep analysis of imaging features and multi-omics data. CT scans and MRI images, as the most commonly used imaging tools, help doctors make judgments by analyzing features such as lesion size and shape. Machine learning can extract characteristics such as tumor morphology, size, density, and texture, allowing it to identify potential lesions even before clear clinical symptoms appear. By analyzing features like irregular tumor borders and vascular invasion, machine learning can alert clinicians to high-risk patients without obvious symptoms, assisting in decisions about whether to proceed with surgery or opt for non-surgical treatment. Additionally, features such as cavitation and cystic changes in imaging can reflect the prognosis of the disease. In conjunction with imaging data, integrated multi-omics analysis (Table 2) can also help identify early biomarkers for HAE. Machine learning can extract features from miRNA1,lncRNAs14, transcriptomics15 and metabolomics , revealing molecular-level changes in the liver of HAE patients. The significant expression of certain miRNAs in HAE, as well as changes in the expression of liver fibrosis-related genes and immune factors, not only aid in early diagnosis but also allow for the detection of treatment responses, helping clinicians assess treatment efficacy and adjust treatment plans (Table 3). Table 2 Application Methods/Techniques Imaging Features U-Net16,SVM Surgical Decision XGBoost,LightGBM,CatBoost Multi-Omics Data SNF17,DeepProg17,FM Table 3 Application Area Traditional Methods Machine Learning Diagnosis Relies on imaging examinations and subjective judgment Automatically extracts imaging features, improving diagnostic accuracy and reducing subjective bias Surgical Decision Relies on the experience and clinical judgment of surgeons Provides objective, data-driven decision support based on extensive clinical data Personalization Lacks personalization Analyzes multi-omics and imaging data to revise treatment plans Early Diagnosis Relies on clinical symptoms Combines imaging and omics data to identify potential lesions early Summary The study by Da-Long Zhu et al. applies machine learning to surgical decision-making in HAE, providing a more objective and accurate approach to disease treatment. By using various feature selection methods and the XGBoost model, the study successfully identified key features from a large amount of clinical data. The use of SHAP analysis enhanced the model's interpretability, increasing clinicians' trust in the machine learning model. These findings play a crucial role in surgical decision-making and provide strong data support for HAE treatment decisions. By combining imaging and multi-omics analysis, machine learning can play a significant role in both surgical and non-surgical treatment of HAE. However, current research still faces challenges, such as small sample sizes and single-source data, which affect the external validity and generalizability of the model. Additionally, there is still a lack of feature extraction and in-depth analysis of imaging omics and multi-omics data. Future research should focus on multi-center, large-sample validation to further improve the model's interpretability and clinical relevance. In conclusion, machine learning demonstrates enormous potential in the surgical and non-surgical treatment of HAE. In the future, it will provide data support and objective analysis for early diagnosis, treatment effectiveness evaluation, and long-term prognosis of HAE patients. Citation 1. Boubaker, G. et al. Regulation of hepatic microRNAs in response to early stage Echinococcus multilocularis egg infection in C57BL/6 mice. PLoS Negl Trop Dis 14, e0007640 (2020). 2. Liu, H. et al. Induced hepatocyte-like cells derived from adipose-derived stem cells alleviates liver injury in mice infected with Echinococcus Multilocularis. Sci Rep 14, 26296 (2024). 3. McManus, D. P., Gray, D. J., Zhang, W. & Yang, Y. Diagnosis, treatment, and management of echinococcosis. BMJ 344, (2012). 4. Tang, T. et al. Interpretable machine learning model for predicting post-hepatectomy liver failure in hepatocellular carcinoma. Sci Rep 15, 15469 (2025). 5. Liang, D. et al. Perspective: Global Burden of Iodine Deficiency: Insights and Projections to 2050 Using XGBoost and SHAP. Adv Nutr 16, 100384 (2025). 6. Wang, Z. et al. Multiclassification of Hepatic Cystic Echinococcosis by Using Multiple Kernel Learning Framework and Ultrasound Images. Ultrasound Med Biol 50, 1034–1044 (2024). 7. Aydin, Y. et al. Relevance of Pulmonary Alveolar Echinococcosis. Arch Bronconeumol (Engl Ed) 56, 779–783 (2020). 8. Ellmann, S. et al. Tumor grade-titude: XGBoost radiomics paves the way for RCC classification. Eur J Radiol 188, 112146 (2025). 9. Yanagawa, R. et al. LightGBM outperforms other machine learning techniques in predicting graft failure after liver transplantation: Creation of a predictive model through large-scale analysis. Clin Transplant 38, e15316 (2024). 10. Han, Y. et al. Early prediction of sepsis associated encephalopathy in elderly ICU patients using machine learning models: a retrospective study based on the MIMIC-IV database. Front Cell Infect Microbiol 15, 1545979 (2025). 11. Jin, Y. et al. Classification of Alzheimer’s disease using robust TabNet neural networks on genetic data. Math Biosci Eng 20, 8358–8374 (2023). 12. Rufener, R. et al. Activity of mefloquine and mefloquine derivatives against Echinococcus multilocularis. Int J Parasitol Drugs Drug Resist 8, 331–340 (2018). 13. Dang, Z. et al. A Pilot Study on Developing Mucosal Vaccine against Alveolar Echinococcosis (AE) Using Recombinant Tetraspanin 3: Vaccine Efficacy and Immunology. PLoS Negl Trop Dis 6, e1570 (2012). 14. Nian, X. et al. Understanding pathogen–host interplay by expression profiles of lncRNA and mRNA in the liver of Echinococcus multilocularis-infected mice. PLoS Negl Trop Dis 16, e0010435 (2022). 15. Zhang, X. et al. Transcriptomic Profiling Reveals Gene Expression Changes in Mouse Liver Tissue During Alveolar Echinococcosis. Genes (Basel) 16, 839 (2025). 16. Yousef, R. et al. U-Net-Based Models towards Optimal MR Brain Image Segmentation. Diagnostics (Basel) 13, 1624 (2023). 17. Poirion, O. B., Jing, Z., Chaudhary, K., Huang, S. & Garmire, L. X. DeepProg: an ensemble of deep-learning and machine-learning models for prognosis prediction using multi-omics data. Genome Med 13, 112 (2021).