Using Machine Learning to Unravel the Value of Radiographic Features for the Classification of Bone Tumors

Table 1

Preoperative radiographic features and clinical characteristics with potential clinical importance for diagnosis.

Features	Feature class	Permissible value	ICC
Location	Categorical	Upper tibia (a)/inferior femur (b)/upper humerus (c)/middle humerus (d)	0.954
Location	Categorical	Epiphysis (a)/metaphysis (b)/diaphysis (c)/not applicable (d)	0.854
Eccentric growth	Binary	Without (0)/with (1)	0.921
Expansive growth	Binary	Without (0)/with (1)	0.888
Margin	Binary	Sharp(0)/ill-defined (1)	0.832
Sclerotic border	Binary	Without (0)/with (1)	0.796
Periosteal reaction	Categorical	Without (a)/continuous (b)/interrupted (c)	0.899
Radiographic density	Categorical	Mixed (a)/low (b)/high (c)	0.863
High-density components	Categorical	Without (a)/calcification or ossification (b)/tumor bone (c)/unrecognizable (d)	0.761
Pattern of bone destruction	Categorical	Geographic (a)/moth-eaten (b)/permeated (c)/not applicable (d)	0.812
Source	Binary	Medullary (0)/cortical (1)	0.909
Pathological fracture	Binary	Without (0)/with (1)	0.888
Cortex involvement	Categorical	Complete cortex (a)/cortical expansion and thinning (b)/interrupted cortex (c)	0.870
Clinical data
ESR	Numerical		—
Age	Numerical		—
Gender	Binary	Male (0)/female (1)	—
Redness and hyperemia	Binary	Without (0)/with (1)	—
Swelling	Binary	Without (0)/with (1)	—
Warmth	Binary	Without (0)/with (1)	—
Pain	Binary	Without (0)/with (1)	—
Palpable mass	Binary	Without (0)/with (1)	—
Dyskinesia	Binary	Without (0)/with (1)	—

Note: The location details are shown in supplement section.