Multimodal sono-contrast NIR spectroscopy for breast cancer diagnosis: Cancer characterization modeling and subgroup study

A noninvasive multimodal spectroscopy system was developed for breast cancer diagnosis. It includes three modules: diffuse reflectance spectroscopy (DRS), ultrasonography and low intensity focused ultrasound (LIFU). An IRB approved clinical trial was conducted to evaluate its diagnostic power. Fourteen patients with histopathologically-proven cancer and 26 with benign masses were enrolled with informed consent. Using histopathology as the gold standard, a cancer characterization model was developed to differentiate breast cancer from benign masses. Considering the small dataset, bootstrap resampling method was used to resample the dataset with replacement to create 1000 patient data. Trade-off between sensitivity and specificity was obtained by varying the threshold value in increments of 0.01, from which the receiver-operating characteristic (ROC) curve was generated. The area under the curve (AUC) was calculated to be 0.87. With threshold RMAD 2.28, sensitivity of 84.6% and specificity of 88.0% can be achieved. A sub-group study was also performed to investigate the detection capability of the system for different racial groups. By characterizing the patient population into different racial groups, the diagnostic power of the system can be improved. Current clinical results demonstrated the effectiveness of this promising technique in characterizing cancer from the benign masses. Combining this system with breast ultrasound has the potential to increase the specificity of sonographic breast cancer detection, and to reduce unnecessary invasive procedures.