DeepBT implements a unique dual-channel U-net deep convolutional neural network to automatically segment brain tumors, including vestibular schwannoma, meningiomabrain metastasis. We overcome the issues of different axialplanar imaging resolution, uneven number of different tumors, various sizelocation of tumors, pre-post-radiosurgery tumor changesimage alignment. DeepBT provides a solution to objectively measure tumor volume changes across multiple time points. Furthermore, quantitative radiomic features are extracted from tumor regions in pre-treatment MRI to early predict treatment responseassist clinical decision making.

DeepBT is built based on a large database, including pre-post-treatment data of vestibular schwannoma, meningiomabrain metastasis, to automatically segment tumor lesions with a dice coefficient of 0.8. This technique provides a more reliable measure of volume change across time points,hence alleviates human bias to provide an accurate assessment of therapeutic effects. Through radiomics analysis, we can further predict the treatment response (volume reductionoverall survival)side effects (pseudo-progressionbrain edema) with AUC  0.80 before treatment to promote the development of precision medicine.

"The applicationsvalues of this product are as follows:
 
 1. Help solving the problem of insufficient manpower of hospital doctors by providing reliableaccurate automatic tumor annotation.
 
 2. Accurately identify the extent of tumors whose shapes have changed after treatment.
 
 3. Efficientobjective quantification of volumes at multiple time points, eliminating subjective variability among physicians.
 
 4. Early predict the treatment response to radiosurgeryoptimize treatment strategy. 
 
 5. Improve physician efficiency per unit of time, increase patient trustoutpatient rates, avoid medical dispute costs,use treatment resources more efficiently."