Detection of coxarthrosis in femoral head radiographic images seems limited mainly to vertically oriented pattern features.

Out of 120 conventional hip joint X-rays, two indepenendent examiners have chosen 27 healthy and 62 coxarthrotic joints. Central parts of femoral head images were digitalized (300 points/inch) and pixel density values analysed. Two methods were applied separately to horizontal rows and to vertical columns: variance coefficient calculation and power coefficients of Fourier harmonics. The arithmetic mean and median of variance coefficient for 256 pixel columns were both significantly higher in data of osteoarthrotic femurs (Mann-Whitney U-test, p = 0.0046 and p = 0.0011, respectively), while no difference was found for horizontal rows. The arithmetic mean and median of variance coefficient for 128 pixels long columns were significantly lower in data of osteoarthrotic femurs (p < 0.001) with wider standard deviation (p = 0.0274), while standard deviation was significantly lower in rows of coxarthrotic heads (p < 0.001). Fourier analysis of 128 pixel vertical columns showed significantly higher values in coxarthrotic femoral heads (from 1st harmonic, wave length of 10.8 mm to 33rd harmonic, wave length of 0.328 mm, p < 0.05). Fourier analysis of 128 pixel horizontal rows did not differ much between coxarthrotic and normal femoral heads. Only valuds for the 60th and 61st harmonic (wavelength near 0.2 mm) showed significantly lower power in coxarthrotic images than in controls (p < 0.01). Results suggest that in the analyzed set of digitalized x-ray femoral head images, information regarding osteoarthrotic changes in the central part of femoral head is detectable mainly through mathematic postprocessing of vertically oriented patterns.