FIGURE 3.

(A) LASSO cross-validation error vs. log(Lambda) this sub-figure depicts the relationship between the mean cross-validation (CV) error and the logarithm of the regularization parameter λ in the LASSO regression. The x-axis represents log(λ) ranging from – 8 to 7, and the y-axis shows the mean CV error. The minimum mean CV error occurs at λminλmin, and 1seλ is the largest λ within one standard error of the minimum error. This analysis helps in determining the optimal λ that balances model complexity and prediction error. (B) Log lambda vs. coefficients here, we present how the coefficients of the variables change as log(λ) varies. As log(λ) increases, the coefficients of many variables shrink toward zero. The number of non-zero coefficients decreases with increasing log(λ), demonstrating the variable selection property of the LASSO method. (C) Selected features and their clinical implications the LASSO regression selected “solidity” and “S_mean” as relevant image features for further analysis. Other features such as R_std (standard deviation of red - channel pixel values), B_std (standard deviation of blue - channel pixel values), and B_kurtosis (kurtosis of the blue - channel) were also considered in the initial analysis. R_std and B_std reflect the dispersion of red and blue intensities, respectively, and higher values may indicate an active or proliferative scar. B_kurtosis describes the sharpness of the blue - channel brightness distribution and can help identify abnormal scar patterns related to high - density tissue or calcification.