| 摘要 |
A method for discriminating between benign and malignant tissue lesions and a system configured to carry out the method are disclosed. The system comprises an optical transfer diagnosis device. The method comprising using a plurality of maps of physiology parameters and a plurality of maps of morphology parameters generated from reflectance measurements made from a target tissue lesion. The maps are based on established absorption and transmission spectra from known skin chromophores and mathematical modelling of skin reflectance. A plurality of pure morphology parameters generated from reflectance measurements are also used in the method. The parameters are based on a spatial distribution of the reflectance measurements within the target tissue lesion image at selected wavelengths. Entropies of each of the plurality of maps are calculated as the sum of the mapped parameter multiplied by a logarithm of the parameter and integrated over the area of the target tissue lesion. Cross entropies of the plurality of maps are calculated as the correlation of the calculated entropies among every pair of the plurality of maps. Further calculated are a plurality of pure morphology parameters and a weight is assigned to each the entropy. Further a weight is assigned to the logarithm of each the entropy. A weight is assigned to each the cross entropy and a weight is assigned to the a logarithm of each the cross entropy. Furhter a weight is assigned to each the pure morphology parameter and a weight is assigned to a logarithm of each the pure morphology parameter such that the assigned weights define a vector W and the parameters to which the weights are assigned define a vector P. A diagnostic index D is computed as WTP for a set of lesions known from pathology investigations to be malignant and the average value ? of diagnostic indices is calculated for a set of lesions known from pathology investigations to be benign. A cost function J(W) is defined comprising differences between D and ? and a threshold value V is defined for a diagnostic index based on a set of lesions with known pathology results such that D > V for tissue lesions known to be malignant and D < V for lesions known to be benign. The method further comprises solving an optimization problem by iteratively adjusting the assigned weights until a number of false negatives is zero, a number of false positives is minimized, and J(W) is minimized to determine a weight vector Wm that is derived from the initial weight vector W with assigned weights so as to maximize specificity for 100% sensitivity. The method further comprises calculating a diagnostic index DT = WmTP for the target tissue lesion; and comparing the diagnostic index DT of the target tissue lesion to the threshold value V to determine whether it is benign or malignant. |
