Annotation-efficient cancer detection with report-guided lesion annotation for deep learning-based prostate cancer detection in bpMRI

Abstract

Deep learning-based diagnostic performance increases with more annotated data, but manual annotation is a bottleneck in most fields. Experts evaluate diagnostic images during clinical routine, and write their findings in reports. Automatic annotation based on clinical reports could overcome the manual labelling bottleneck. We hypothesise that dense annotations for detection tasks can be generated using model predictions, guided by sparse information from these reports. To demonstrate efficacy, we generated clinically significant prostate cancer (csPCa) annotations, guided by the number of clinically significant findings in the radiology reports. We included 7,756 prostate MRI examinations, of which 3,050 were manually annotated and 4,706 were automatically annotated. We evaluated the automatic annotation quality on the manually annotated subset: our score extraction correctly identified the number of csPCa lesions for 99.3% of the reports and our csPCa segmentation model correctly localised 83.8 ± 1.1% of the lesions. We evaluated prostate cancer detection performance on 300 exams from an external centre with histopathologyconfirmed ground truth. Augmenting the training set with automatically labelled exams improved patient-based diagnostic area under the receiver operating characteristic curve from 88.1 ± 1.1% to 89.8 ± 1.0% (P = 1.2 • 10 −4 ) and improved lesion-based sensitivity at one false positive per case from 79.2 ± 2.8% to 85.4 ± 1.9% (P < 10 −4 ), with mean ± std. over 15 independent runs. This improved performance demonstrates the feasibility of our reportguided automatic annotations. Source code is made publicly available at github.com/DIAGNijmegen/Report-Guided-Annotation. Best csPCa detection algorithm is made available at grand-challenge.org/algorithms/bpmri-cspcadetection-report-guided-annotations/

Prostate Lesion Estimation using Prostate Masks from Biparametric MRI

“…ADC images were normalized with complete z-score normalization with respect to the entire dataset while T2 and DWI images were normalized with instancewise z-score normalization since ADC images are more robust than T2 and DWI images. Lesion segmentation was carried out using the nnU-Net framework, but instead of the default combination of dice and cross entropy loss, Focal Loss and cross entropy loss were employed, as in [2].…”

“…Additionally, test-time augmentation was used to improve the model's performance by making predictions on many augmented images. To create a detection map that is required by the PICAI evaluation step, we acquired unique lesion candidates, as done in [2], using the voxel-level confidence maps that were produced. By starting at the voxel with the highest degree of confidence and encompassing all related voxels (in 3D) with at least 40% of the peak's degree of confidence, we specifically produced a lesion candidate.…”

Artificial intelligence for prostate MRI: open datasets, available applications, and grand challenges

“…Patient age, PSA density, PSA level and prostate volume will be provided for all cases. Expert-derived lesion delineations are provided for approximately 80% of all cases, and AI-derived lesion delineations (pseudo-labels) are provided for all cases, using a state-of-the-art csPCa detection developed at Radboudumc [48].…”

Equilibrium Optimization Algorithm with Deep Learning Enabled Prostate Cancer Detection on MRI Images

“…In several countries, the standard practice for PrC diagnosis relies on high rates of prostate-specific antigen (PSA) in the blood and a digital rectal examination (DRE). In some cases, pre-biopsy magnetic resonance imaging (MRI) may be recommended to guide the biopsy process [2]. Automated computer-aided diagnosis (CAD) and identification systems can address the limitations of the standard radiological analysis by applying quantitative techniques for automated, standardized, and supported regenerative analyses of radiological images [3].…”