Artificial Intelligence-Aided Detection Models for Diagnostic Imaging in Prostate Cancer
Abstract
Prostate cancers (CaPs) exhibit diverse clinical behaviors, ranging from slow-growing to aggressive disease that develops metastases rapidly and causes significant illness and death. A vital clinical need is to identify characteristics that distinguish slow-growing tumors from aggressive tumors for the purpose of directing treatments to the latter. Nowadays, multiparametric Magnet Resonance Imaging of Prostate (mpMRI) plays an essential role in the detection of significant CaPs, which have Gleason scores higher than 6, but it is still limited by the dependency on the subjective interpretation. A Gleason score is a scoring system for microscopic tumor appearance associated with tumor progression. Developing computer-aided diagnostic tools for MR imaging using intelligent algorithms could greatly contribute to the accuracy of the tumor detection and localization and potentially serve as a guide for localized therapy. Computer-aided diagnostic tools can further improve the detection of significant CaP by lowering the false-positive rates. Hence, the objective of this study is to develop computer-aided diagnostic and prognostic tools that identify and localize significant CaP lesions from features found on MRI and histology images. To this end, I plan to evaluate this objective by pursuing the following two Specific Aims: Aim 1. To develop image processing methods that can determine significant tumor lesions on mpMRI patient-specific feature maps that will be generated from mpMRI for 428 cases with CaP who were treated with total removal of the prostate. Further, tumors lesions found on histology images and mpMRI will be delineated and annotated with the significant tumor status and the Gleason score grading system. After data and image processing, these feature maps will be stored with the corresponding histology images, including histopathology results from the removed prostate, using a file system called cMDX that enables the integration of multiple datasets into a single file, thereby simplifying the data access. I will evaluate various intelligent algorithms that predict the significant tumor lesions found on the removed prostate from their MRI features profiles. I will compare the accuracy of these models to determine the best model for the detection of significant lesions on the mpMRI. I will further implement the best model in the picture archiving and communication system at a single institution to evaluate its clinical feasibility. Aim 2. To identify tumor lesions with the risk for CaP recurrence Histology images from 400 CaP cases who were treated with the complete removal of the prostate will be randomly selected from the McNeal s dataset, which includes data from patients followed for a long time over 10 years. A subset of the study cases will be used to develop and evaluate a tumor detection approach that uses intelligent algorithms and enables automatic registration of the tumor lesions on all histology images, thereby reducing the annotation duration. Our pathologists will then evaluate the detection accuracy. After that, I am going to develop and evaluate a prediction tool for CaP with high chances of coming back after the total removal of the prostate using intelligent algorithms and histology images. The proposed project will provide a valuable training opportunity to further my experience in clinical imaging research and artificial intelligence, become an expert in these fields, and build networks with well-recognized institutions and researchers. During this project, I will be attending and auditing courses at Stanford University to acquire the skill, competence, and recent knowledge of artificial intelligence and to build a connection with lecturers to receive their feedback regarding my project. Further, I will interact with leaders in the CaP field to receive their expert opinions on the early detection of CaP and diagnostic imaging. I will also interact with experts in MRI, arti
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 29, 2018
- Source ID
- W81XWH1810396
Entities
People
- Okyaz Eminaga
Organizations
- Stanford University
- United States Army