ABSTRACT
Objective.Deep learning models that aid in medical image assessment tasks must be both accurate and reliable to be deployed within clinical settings. While deep learning models have been shown to be highly accurate across a variety of tasks, measures that indicate the reliability of these models are less established. Increasingly, uncertainty quantification (UQ) methods are being introduced to inform users on the reliability of model outputs. However, most existing methods cannot be augmented to previously validated models because they are not post hoc, and they change a model's output. In this work, we overcome these limitations by introducing a novel post hoc UQ method, termedLocal Gradients UQ, and demonstrate its utility for deep learning-based metastatic disease delineation.Approach.This method leverages a trained model's localized gradient space to assess sensitivities to trained model parameters. We compared the Local Gradients UQ method to non-gradient measures defined using model probability outputs. The performance of each uncertainty measure was assessed in four clinically relevant experiments: (1) response to artificially degraded image quality, (2) comparison between matched high- and low-quality clinical images, (3) false positive (FP) filtering, and (4) correspondence with physician-rated disease likelihood.Main results.(1) Response to artificially degraded image quality was enhanced by the Local Gradients UQ method, where the median percent difference between matching lesions in non-degraded and most degraded images was consistently higher for the Local Gradients uncertainty measure than the non-gradient uncertainty measures (e.g. 62.35% vs. 2.16% for additive Gaussian noise). (2) The Local Gradients UQ measure responded better to high- and low-quality clinical images (p< 0.05 vsp> 0.1 for both non-gradient uncertainty measures). (3) FP filtering performance was enhanced by the Local Gradients UQ method when compared to the non-gradient methods, increasing the area under the receiver operating characteristic curve (ROC AUC) by 20.1% and decreasing the false positive rate by 26%. (4) The Local Gradients UQ method also showed more favorable correspondence with physician-rated likelihood for malignant lesions by increasing ROC AUC for correspondence with physician-rated disease likelihood by 16.2%.Significance. In summary, this work introduces and validates a novel gradient-based UQ method for deep learning-based medical image assessments to enhance user trust when using deployed clinical models.
Subject(s)
Deep Learning , Image Processing, Computer-Assisted , Uncertainty , Humans , Image Processing, Computer-Assisted/methodsABSTRACT
Objective.Fast and accurate deformable image registration (DIR), including DIR uncertainty estimation, is essential for safe and reliable clinical deployment. While recent deep learning models have shown promise in predicting DIR with its uncertainty, challenges persist in proper uncertainty evaluation and hyperparameter optimization for these methods. This work aims to develop and evaluate a model that can perform fast DIR and predict its uncertainty in seconds.Approach.This study introduces a novel probabilistic multi-resolution image registration model utilizing convolutional neural networks to estimate a multivariate normal distributed dense displacement field (DDF) in a multimodal image registration problem. To assess the quality of the DDF distribution predicted by the model, we propose a new metric based on the Kullback-Leibler divergence. The performance of our approach was evaluated against three other DIR algorithms (VoxelMorph, Monte Carlo dropout, and Monte Carlo B-spline) capable of predicting uncertainty. The evaluation of the models included not only the quality of the deformation but also the reliability of the estimated uncertainty. Our application investigated the registration of a treatment planning computed tomography (CT) to follow-up cone beam CT for daily adaptive radiotherapy.Main results.The hyperparameter tuning of the models showed a trade-off between the estimated uncertainty's reliability and the deformation's accuracy. In the optimal trade-off, our model excelled in contour propagation and uncertainty estimation (p <0.05) compared to existing uncertainty estimation models. We obtained an average dice similarity coefficient of 0.89 and a KL-divergence of 0.15.Significance.By addressing challenges in DIR uncertainty estimation and evaluation, our work showed that both the DIR and its uncertainty can be reliably predicted, paving the way for safe deployment in a clinical environment.
Subject(s)
Image Processing, Computer-Assisted , Neural Networks, Computer , Uncertainty , Image Processing, Computer-Assisted/methods , Humans , Algorithms , Radiotherapy Planning, Computer-Assisted/methods , Cone-Beam Computed Tomography/methodsABSTRACT
Objective. Manual analysis of individual cancer lesions to assess disease response is clinically impractical and requires automated lesion tracking methodologies. However, no methodology has been developed for whole-body individual lesion tracking, across an arbitrary number of scans, and acquired with various imaging modalities.Approach. This study introduces a lesion tracking methodology and benchmarked it using 2368Ga-DOTATATE PET/CT and PET/MR images of eight neuroendocrine tumor patients. The methodology consists of six steps: (1) alignment of multiple scans via image registration, (2) body-part labeling, (3) automatic lesion-wise dilation, (4) clustering of lesions based on local lesion shape metrics, (5) assignment of lesion tracks, and (6) output of a lesion graph. Registration performance was evaluated via landmark distance, lesion matching accuracy was evaluated between each image pair, and lesion tracking accuracy was evaluated via identical track ratio. Sensitivity studies were performed to evaluate the impact of lesion dilation (fixed versus automatic dilation), anatomic location, image modalities (inter- versus intra-modality), registration mode (direct versus indirect registration), and track size (number of time-points and lesions) on lesion matching and tracking performance.Main results. Manual contouring yielded 956 lesions, 1570 lesion-matching decisions, and 493 lesion tracks. The median residual registration error was 2.5 mm. The automatic lesion dilation led to 0.90 overall lesion matching accuracy, and an 88% identical track ratio. The methodology is robust regarding anatomic locations, image modalities, and registration modes. The number of scans had a moderate negative impact on the identical track ratio (94% for 2 scans, 91% for 3 scans, and 81% for 4 scans). The number of lesions substantially impacted the identical track ratio (93% for 2 nodes versus 54% for ≥5 nodes).Significance. The developed methodology resulted in high lesion-matching accuracy and enables automated lesion tracking in PET/CT and PET/MR.
Subject(s)
Neuroendocrine Tumors , Positron Emission Tomography Computed Tomography , Humans , Tomography, X-Ray Computed/methods , Multimodal Imaging/methods , Positron-Emission Tomography/methods , Neuroendocrine Tumors/diagnostic imaging , Magnetic Resonance Imaging/methodsABSTRACT
OBJECTIVE: H3K27M mutation in gliomas has prognostic implications. Previous magnetic resonance imaging (MRI) studies have reported variable rates of tumoral enhancement, necrotic changes, and peritumoral edema in H3K27M-mutant gliomas, with no distinguishing imaging features compared with wild-type gliomas. We aimed to construct an MRI machine learning (ML)-based radiomic model to predict H3K27M mutation in midline gliomas. METHODS: A total of 109 patients from 3 academic centers were included in this study. Fifty patients had H3K27M mutation and 59 were wild-type. Conventional MRI sequences (T1-weighted, T2-weighted, T2-fluid-attenuated inversion recovery, postcontrast T1-weighted, and apparent diffusion coefficient maps) were used for feature extraction. A total of 651 radiomic features per each sequence were extracted. Patients were randomly selected with a 7:3 ratio to create training (n = 76) and test (n = 33) data sets. An extreme gradient boosting algorithm (XGBoost) was used in ML-based model development. Performance of the model was assessed by area under the receiver operating characteristic curve. RESULTS: Pediatric patients accounted for a larger proportion of the study cohort (60 pediatric [55%] vs. 49 adult [45%] patients). XGBoost with additional feature selection had an area under the receiver operating characteristic curve of 0.791 and 0.737 in the training and test data sets, respectively. The model achieved accuracy, precision (positive predictive value), recall (sensitivity), and F1 (harmonic mean of precision and recall) measures of 72.7%, 76.5%, 72.2%, and 74.3%, respectively, in the test set. CONCLUSIONS: Our multi-institutional study suggests that ML-based radiomic analysis of multiparametric MRI can be a promising noninvasive technique to predict H3K27M mutation status in midline gliomas.