Abstract 2103: Detecting pancreatic cancer through neural network analysis of human plasma MR spectra

Abstract

Abstract The poor prognosis of pancreatic ductal adenocarcinoma (PDAC) is mainly due to late-stage diagnosis [1]. Similarities in the clinical behavior and imaging features of PDAC and chronic pancreatitis further complicate the detection of PDAC [2]. There is an urgent need for noninvasive clinically translatable biomarkers of PDAC. Plasma based tests provide an attractive option for routine screening. Here we have applied neural network analysis to 1H magnetic resonance (MR) spectra of human plasma samples to differentiate between healthy subjects (control), subjects with benign lesions, and subjects with pancreatic ductal adenocarcinoma (PDAC). Our data support developing a neural-network approach to identify PDAC from 1H MRS of plasma samples. Plasma samples from healthy subjects (control, n=28), from subjects with benign pancreatic lesions (benign, n=32), and from subjects with PDAC (PDAC/malignant, n=34) were analyzed with 1H MRS. 1H MR spectra were acquired on a Bruker Avance III 750 MHz (17.6 T) MR spectrometer equipped with a 5 mm probe. After processing the spectral data for normalization against a reference peak signal and plasma volume, spectral features were extracted to develop an artificial neural-network technique to discriminate between the three classes of spectra. Three spectral features showed high probability for successful discrimination: (a) signal intensity weighted-mean of the spectra, (b) standard deviation of the spectra about the weighted-mean from (a), and (c) weighted mean of the difference spectra. This led to the design of our three-layer artificial neural network model to solve the discrimination problem. The artificial neural-network, developed in MATLAB 2019b (MathWorks, Inc), was limited to a stack of two auto-encoder layers and a ‘softmax' layer. Due to the limited size of the data all of the data were used to train the network while keeping low L2-regularization to avoid over-training. Cross-validation results were captured as confusion matrices and receiver operating characteristic (ROC) curves. With a combination of spectral features extraction and neural network processing of MRS data of plasma samples we could discriminate between control, benign and PDAC. The sensitivity and specificity for detecting PDAC was 84% and 95%, respectively. Our data suggest that the approach can be extended to encompass a large sample size to improve accuracy and provide a rigorous and robust solution for plasma based detection of PDAC. Acknowledgement: This work was supported by NIH R35CA209960 and R01CA193365.