Purpose. To evaluate utility of 3D model texture analysis of the pathological process for morphological diagnostics of localized renal parenchymal neoplasms.

Material and methods. Retrospective research was performed, including 50 patients with executed laparoscopic renal resections. All cases were divided into 5 similar groups. 3 groups were consisted of patients with malignant tumors: clear cell renal cell carcinoma (RCC) G1, papillary RCC type 1 G1 and chromophobe RCC. 2 groups included patients with benign neoplasms: angiomyolipoma and oncocytoma. All patients underwent 3D virtual surgery planning using the 3D modeling program "Amira". 3D models of tumors were undergone to textural analysis. For assessment of tumor histograms were used 4 features of 1st order statistics: mean of gray level intensity, standard deviation of gray level intensity, skewness and kurtosis. 6 textural characteristics of 2nd order statistics: autocorrelation, entropy, uniformity, energy, contrast and sum of squares.

Machine learning algorithms were used to determine morphological types of neoplasms and calculate classification accuracy. The best result was demonstrated by the "Logistic Regression" algorithm, which used first- and second-order statistics as input parameters.

Results. Group of angiomyolipoma had 80% accuracy using autocorrelation and contrast parameters. Group of oncocytoma had 70% accuracy using entropy and kurtosis parameters or entropy and contrast parameters. Group of papillary RCC had 70% accuracy using autocorrelation and sum of squares parameters. Group of clear cell RCC had 50% accuracy using mean of gray level intensity or standard deviation parameter. Chromophobe RCC had 70% accuracy using autocorrelation, contrast, entropy and energy parameters.

Conclusion. The use of textural analysis of 3D models of kidney formations made it possible to morphologically verify malignant tumors of the kidney parenchyma with an accuracy of 50%; 70%; 80% (light cell RCC; chromophobic RCC; papillary RCC) and benign neoplasms with an accuracy of 70%; 80% (oncocytoma, angiomyolipoma).