A Deep Learning Approach For Identifying Focal Prostate Cancer From Multi-Parametric Mri

There are no technical barriers to delivering radiotherapy to small focal lesions within the prostate, however, reliably identifying focal disease is challenging. Multi-parametric magnetic resonance imaging (mp-MRI) has potential for this and because of its improved image resolution it may be combined with machine learning to assist with delineation. The aim of this work was to combine information from T2 weighted, apparent diffusion coefficient (ADC), and diffusion weighted MRI to train machine learning models to identify focal disease within the prostate. Two datasets were utilised from previously treated patients with localised prostate cancer. The first included 16 patients with diagnostic T2 MRI, the second included 12 patients with diagnostic T2 and ADC studies. The planning CT, T2 and ADC images, where available, were registered rigidly and a clinician contoured the prostate and focal lesion on each image. Using MATLAB, sub-images were extracted from each before 32 texture features were calculated and used to train four different classification algorithms. In addition, a pre-trained convolution neural network was fine-tuned to classify each sub-image as healthy or diseased tissue. The performance of each model was assessed in terms of sensitivity, specificity and AUC. Results demonstrate that mp-MRI images can be successfully combined to identify focal disease using machine learning. This novel approach achieved a high classification performance when tested on T2 images with an AUC of 0.935 compared to 0.663 found using single sequence MRI studies. These results are promising, yet a larger data set is required to further develop these approaches.