Joint Prostate Cancer Detection and Gleason Score Prediction in mp-MRI via FocalNet

Abstract

Multi-parametric MRI (mp-MRI) is considered the best non-invasive imaging modality for diagnosing prostate cancer (PCa). However, mp-MRI for PCa diagnosis is currently limited by the qualitative or semi-quantitative interpretation criteria, leading to inter-reader variability and a suboptimal ability to assess lesion aggressiveness. Convolutional neural networks (CNNs) are a powerful method to automatically learn the discriminative features for various tasks, including cancer detection. We propose a novel multi-class CNN, FocalNet, to jointly detect PCa lesions and predict their aggressiveness using Gleason score (GS). FocalNet characterizes lesion aggressiveness and fully utilizes distinctive knowledge from mp-MRI. We collected a prostate mp-MRI dataset from 417 patients who underwent 3T mp-MRI exams prior to robotic-assisted laparoscopic prostatectomy (RALP). FocalNet is trained and evaluated in this large study cohort with 5-fold cross-validation. In the free-response receiver operating characteristics (FROC) analysis for lesion detection, FocalNet achieved 89.7% and 87.9% sensitivity for index lesions and clinically significant lesions at 1 false positive per patient, respectively. For GS classification, evaluated by the receiver operating characteristics (ROC) analysis, FocalNet received the area under the curve (AUC) of 0.81 and 0.79 for the classifications of clinically significant PCa (GS≥3+4) and PCa with GS≥4+3, respectively. With the comparison to the prospective performance of radiologists using the current diagnostic guideline, FocalNet demonstrated comparable detection sensitivity for index lesions and clinically significant lesions, only 3.4% and 1.5% lower than highly experienced radiologists without statistical significance.

Radiologist-like artificial intelligence for grade group prediction of radical prostatectomy for reducing upgrading and downgrading from biopsy

“…In these, attention was paid primarily to GS-7 cases at biopsy core specimens. Unfortunately, few studies have attempted to build a patient-level hierarchical prediction model to support clinical decision-making by addressing up-/downgrading alterations 21 . Notably, deep reinforcement learning (DRL) has an advantage with complex reasoning tasks because of its environmental perception, which could provide an analysis approach from slice-level results to patient-level results 22 .…”

Registration of presurgical MRI and histopathology images from radical prostatectomy via RAPSODI

“…Although numerous automated approaches for the registration of radiology and histopathology images have been proposed previously (see detailed discussion in the Supplementary material), manual registration approaches are still employed, even in recent publications. [9][10][11][12][13] These manual approaches can generate subjective results and are tedious to use. They either rely on the user's expertise to identify and pick corresponding landmarks in the histopathology images and MRI 9,10,13 or use cognitive alignments.…”

“…Although numerous approaches for the radiology‐pathology registration in the prostate have been introduced (see section “Prior Work” in the Suplementary Material), these approaches have not been widely adopted and have not been carefully tested by scientists outside the developer teams. Recent publications using histopathology images as a reference to improve MRI and automatically detect cancer 9–13 still require manual approaches to align the histopathology to MR images; these approaches are labor‐intensive and subjective. The slow adoption of previous methods is due to the challenges associated with managing and registering the histopathology images and MRI, the lack of open source methods, and the time constraints associated with running these methods.…”

A Comparison of Transfer Learning Performance Versus Health Experts in Disease Diagnosis From Medical Imaging

“…The open access series of breast ultrasonic dataset, which contains 882 images of unique breast masses, consists of 678 benign and 204 malignant lesions [23]. Cao et al [58] were practiced prostrate mp-MRI dataset, which contains the data of 417 patients, preprocessed by intensity normalization and 3T scanners were used to take these images. Chee et al [59] were used the hips dataset collected by Seoul National University Hospital (SNUH), which contains 673, 1346 MRI images of 16 years old or older patients and DICOM radiographic image archive were loaded by using python library 0.9.9v.…”

“…Remarkably, the pretrained models were able to identify the NoF with similar levels as radiologic technologist. Cao et al, [58] used pretrained FocalNet technique for the detection joint prostate cancer and calculate Gleason score. For the assessment of their method, they used mp-MRI dataset of 417 patients and achieved sensitivity of 89.7%.…”