Regression model for diagnosis of breast pathology according to microwaves radiometry data

The paper by T.V. Zamechnik, A.G. Losev and E.A. Mazepa [6] set out an algorithm for obtaining highly informative diagnostic signs for breast pathology based on microwave radiometry. This paper examines the effect of the temperature at the reference points on the informative features. Obviously, the results of measurements depend on the ambient temperature. Unfortunately, the ambient temperature was not recorded during creation of the training sample. For analysis of indirect effects of ambient temperature we decided to use the temperature at control points $T1$ and $T2$. Analysis of the corresponding correlation coefficients revealed that the temperature at the control points has high direct correlation with the temperature changes in the mammary glands.
Learning sample data was pre-processed. We obtained linear regressions depending o the results of measurements of the temperature at the control points. Thereafter, we reduced the measurement results to an average temperature at reference points $T1$ and $T2$. The preprocessing of sample data resulted in increasing efficiency of some characteristic features for diagnosis and improved the information content of highly informative signs. So it improved the accuracy of the diagnostic algorithm.
The paper attempts to use non-linear regression models which can be linearized. For the new training samples we used hyperbolic, logarithmic, power and exponential regressions. Using these types of regression doesn't give any new results because the regression lines of all kinds are almost identical within a range of patients temperature.