Related Experiment Video
Updated: Jun 2, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Radiomics-Based Machine Learning versus the MAP Score for Predicting Haematocrit Decline after Laparoscopic Partial
Yusuf Dogan1, Ahmet Alper Ozdes2, Mustafa Yildirim3
1Department of Radiology, Etlik City Hospital, Ankara, Turkey.
Objective:
To evaluate and compare the ability of the Mayo Adhesive Probability (MAP) score and radiomics-based machine learning approaches to predict the degree of postoperative haematocrit (Hct) reduction using preoperative computed tomography (CT) images in patients with renal cell carcinoma (RCC) treated with laparoscopic partial nephrectomy (LPN).
Study Design:
An observational retrospective study. Place and Duration of the Study: Departments of Urology and Radiology, Fırat University Hospital, Elazig, Turkey, from 2020 to 2025.
Methodology:
Eligible Patients were categorised into high (>6 units) and low Hct decline groups. MAP scores were assigned by an experienced radiologist based on preoperative contrast-enhanced CT findings. A total of 107 radiomics features were extracted from the tumour and the adjacent perirenal fat and supplied to seven machine-learning classifiers (SVM, logistic regression, kNN, Naive Bayes, tree, random forest, and neural network) using 10‑fold cross‑validation for training and validation. The models' predictive ability was quantified by the area under the receiver operating characteristic curve (AUC), with sensitivity and specificity reported as additional performance metrics.
Results:
The study population included 53 patients (34 men and 19 women; mean age: 60.38 ± 10.82 years). The MAP score at a 1.5 cut-off yielded an AUC of 0.777 (95% CI: 0.622-0.933; p = 0.002), a sensitivity of 71.8%, and a specificity of 85.7% in predicting high Hct decline. The radiomics-based Naive Bayes model demonstrated strong predictive performance, yielding an AUC of 0.925, an overall accuracy of 0.950, a sensitivity of 90%, and a specificity of 100%.
Conclusion:
Radiomics analysis combined with machine learning outperformed the MAP adhesive score in predicting postoperative Hct decline after LPN for RCC. The integration of radiomics into preoperative assessment may optimise surgical planning, improve patient safety, and reduce the risk of complications. Multicentre prospective studies are needed before transitioning to clinical practice.
Key Words:
Radiomics, Mayo adhesive probability, Renal cell carcinoma, Laparoscopic partial nephrectomy.
