Peritumoral features have been suggested to be useful in improving the prediction performance of radiomic models.
Our lab has conducted a study in order to systematically investigate the prediction performance improvement for sentinel lymph node (SLN) status in breast cancer from peritumoral features in radiomic analysis by exploring the effect of peritumoral region sizes.
Findings:
The accuracy in the validation set using the two radiologists’ ROIs improved from 0.704 to 0.796 by incorporating peritumoral features.
The choice of the peritumoral size could affect the level of improvement.
A phantom study has been performed to systematically evaluate the behavior of radiomic features under various conditions (signal to noise ratio, region of interest delineation, voxel size change and normalization methods) using intraclass correlation coefficients. The features extracted in this phantom study include first order, shape, gray level cooccurrence matrix and gray level run length matrix. Many features are found to be non-robust to changing parameters.
E.g.
Image of (a) regions of interest under investigation in this study, namely pineapple core (red), banana (blue), orange (orange) and kiwi (green), and (b) regions of interest used for signal to noise ratio calculation
Magnitude images at different signal to noise ratio (SNR) steps: (a) SNR = 45, (b) SNR = 75 and (c) SNR = 124
Ki-67 is one of the most important biomarkers of breast cancer traditionally measured invasively via immuno-histochemistry.
In this study, we have established a preoperative Ki-67 classification model in patients with breast cancer using transfer learn- ing from deep CNN, based on T2WI, DWI, T1+C images, and mp-MRI. The analysis revealed that our deep learning model, especially the mp-MRI based model, had the ability to classify a patient's Ki-67 status noninvasively with good classification performance
Deep learning workflow. (1) Image preprocessing mainly included normalization and channel expansion. (2) Transfer VGG was mainly used for feature extraction. (3) Maximum relevance was used for feature selection. (4) Multi-layer perception (MLP) classifiers were con- structed.
ROC curves of the four deep learning models on the training datasets. The AUCs in the mp-MRI (green solid line),T1+C (yellow solid line), T2WI (red solid line), DWI (blue solid line) were 0.888 (95% CI: 0.881,0.895), 0.873 (95% CI: 0.863,0.883), 0.727 (95% CI: 0.721,0.733) and 0.674 (95% CI: 0.665,0.684), respectively.
ROC curves of the four deep learning models on the validation datasets. The AUCs in the mp-MRI (green solid line), T1+C (yellow solid line), T2WI (red solid line), DWI (blue solid line) were 0.875 (95% CI: 0.871,0.880), 0.829 (95% CI: 0.824,0.834), 0.706 (95% CI: 0.701,0.711) and 0.643 (95% CI: 0.629,0.657), respectively.