A method for adjusting directed texture features in biomedical image analysis problems

As part of the general problem of automatic information feature construction, we considered a particular applied problem of the calculation direction adjustment for the directed texture features intended to diagnose various diseases from digital biomedical images. As feature space quality criteria, we considered the classification accuracy, Bhattacharyya distance and the discriminant analysis criteria. We used random search, a genetic algorithm and simulated annealing as the optimization algorithms. The proposed approach enables a two-fold reduction in the error probability estimation when diagnosing bone tissue X-ray images (from 0.20 to 0.10), also enabling a 45-percent error reduction when diagnosing computed tomography (CT) lung images (from 0.11 to 0.06) in comparison with conventional procedures of selecting from a large number of heterogeneous features.