ABSTRACT
Accurate and early identification of renal rejection type post-transplantation is crucial to provide the proper management plan. Biopsy, the gold standard, is invasive, expensive, and time consuming. Thus, there is an urgent need for a new technology that can non-invasively and precisely specify the type of renal rejection after transplantation at an early stage. Here, a new computer-aided diagnostic (CAD) is presented with the ability to classify renal rejection types by utilizing analyzing apparent diffusion coefficients (ADCs). These ADC maps are constructed from diffusion-weighted magnetic resonance imaging (DW-MRI) data, which have been obtained at low and high b-values to account for both perfusion and diffusion. The proposed CAD system mainly performs the following steps to get the final diagnosis: (1) pre-processing of the DW-MRI data by handling the noise effects and the motion that might occur during data acquisition followed by an accurate segmentation of the renal allograft using the level-sets technique for further analysis; (2) extraction of the distinguishing features represented by the cumulative distribution functions (CDFs) of the ADCs at both high and low b-values; and (3) a deep learning diagnosis of the renal allograft as acute tubular necrosis (ATN) or T-cell mediated rejection (TMR) based on using a stacked autoencoder with an additional non-negativity constraint for a faster solution divergence. By using a leave-one-out cross-validation (LOOCV) approach on 39 biopsy-proven renal allografts (ATN = 8 and TMR = 31), the presented CAD system achieved an accuracy of 98% with 8/8 ATN and 30/31 TMR correctly classified instances. These results demonstrated the feasibility and efficacy of the proposed CAD system to be reliably used to identify the renal allograft rejection type in a non-invasive way.
