Use of texture parameters in the classification of soft tissue tumors

Soft tissue tumors are a challenging group of tumors that presents wide morphologic variety. For pathologists, the degree of cellular differentiation, the cellular and nuclear polymorphism and the mitotic rate are major criteria for diagnosis. In image analysis, the architectural organization of such tumors shows local variations that can be understood as texture variations. This study introduced a method of measuring the architectural organization of soft tissue tumors from texture analysis of tissue sections at low magnification. We studied 40 cases of soft tissue tumors classified by pathologists into three groups according to their histologic patterns. Twelve texture parameters were calculated on subimages of 128 x 128 pixels. Our results show that tissue architecture evaluated by texture analysis provides good discrimination of myxoid, spindle cell and round cell tumors.