Development and evaluation of intraoperative ultrasound segmentation with negative image frames and multiple observer labels

When developing deep neural networks for segmenting intraoperative ultrasound images, several practical issues are encountered frequently, such as the presence of ultrasound frames that do not contain regions of interest and the high variance in ground-truth labels. In this study, we evaluate the utility of a pre-screening classification network prior to the segmentation network. Experimental results demonstrate that such a classifier, minimising frame classification errors, was able to directly impact the number of false positive and false negative frames. Importantly, the segmentation accuracy on the classifier-selected frames, that would be segmented, remains comparable to or better than those from standalone segmentation networks. Interestingly, the efficacy of the pre-screening classifier was affected by the sampling methods for training labels from multiple observers, a seemingly independent problem. We show experimentally that a previously proposed approach, combining random sampling and consensus labels, may need to be adapted to perform well in our application. Furthermore, this work aims to share practical experience in developing a machine learning application that assists highly variable interventional imaging for prostate cancer patients, to present robust and reproducible open-source implementations, and to report a set of comprehensive results and analysis comparing these practical, yet important, options in a real-world clinical application.

Controlling False Positive/Negative Rates for Deep-Learning-Based Prostate Cancer Detection on Multiparametric MR images

Prostate cancer (PCa) is one of the leading causes of death for men worldwide. Multi-parametric magnetic resonance (mpMR) imaging has emerged as a non-invasive diagnostic tool for detecting and localising prostate tumours by specialised radiologists. These radiological examinations, for example, for differentiating malignant lesions from benign prostatic hyperplasia in transition zones and for defining the boundaries of clinically significant cancer, remain challenging and highly skill-and-experience-dependent. We first investigate experimental results in developing object detection neural networks that are trained to predict the radiological assessment, using these high-variance labels. We further argue that such a computer-assisted diagnosis (CAD) system needs to have the ability to control the false-positive rate (FPR) or false-negative rate (FNR), in order to be usefully deployed in a clinical workflow, informing clinical decisions without further human intervention. This work proposes a novel PCa detection network that incorporates a lesion-level cost-sensitive loss and an additional slice-level loss based on a lesion-to-slice mapping function, to manage the lesion- and slice-level costs, respectively. Our experiments based on 290 clinical patients concludes that 1) The lesion-level FNR was effectively reduced from 0.19 to 0.10 and the lesion-level FPR was reduced from 1.03 to 0.66 by changing the lesion-level cost; 2) The slice-level FNR was reduced from 0.19 to 0.00 by taking into account the slice-level cost; (3) Both lesion-level and slice-level FNRs were reduced with lower FP/FPR by changing the lesion-level or slice-level costs, compared with post-training threshold adjustment using networks without the proposed cost-aware training.

Learning to Address Intra-segment Misclassification in Retinal Imaging

Accurate multi-class segmentation is a long-standing challenge in medical imaging, especially in scenarios where classes share strong similarity. Segmenting retinal blood vessels in retinal photographs is one such scenario, in which arteries and veins need to be identified and differentiated from each other and from the background. Intra-segment misclassification, i.e. veins classified as arteries or vice versa, frequently occurs when arteries and veins intersect, whereas in binary retinal vessel segmentation, error rates are much lower. We thus propose a new approach that decomposes multi-class segmentation into multiple binary, followed by a binary-to-multi-class fusion network. The network merges representations of artery, vein, and multi-class feature maps, each of which are supervised by expert vessel annotation in adversarial training. A skip-connection based merging process explicitly maintains class-specific gradients to avoid gradient vanishing in deep layers, to favor the discriminative features. The results show that, our model respectively improves F1-score by 4.4\%, 5.1\%, and 4.2\% compared with three state-of-the-art deep learning based methods on DRIVE-AV, LES-AV, and HRF-AV data sets.

Learning image quality assessment by reinforcing task amenable data selection

In this paper, we consider a type of image quality assessment as a task-specific measurement, which can be used to select images that are more amenable to a given target task, such as image classification or segmentation. We propose to train simultaneously two neural networks for image selection and a target task using reinforcement learning. A controller network learns an image selection policy by maximising an accumulated reward based on the target task performance on the controller-selected validation set, whilst the target task predictor is optimised using the training set. The trained controller is therefore able to reject those images that lead to poor accuracy in the target task. In this work, we show that the controller-predicted image quality can be significantly different from the task-specific image quality labels that are manually defined by humans. Furthermore, we demonstrate that it is possible to learn effective image quality assessment without using a ``clean'' validation set, thereby avoiding the requirement for human labelling of images with respect to their amenability for the task. Using $6712$, labelled and segmented, clinical ultrasound images from $259$ patients, experimental results on holdout data show that the proposed image quality assessment achieved a mean classification accuracy of $0.94\pm0.01$ and a mean segmentation Dice of $0.89\pm0.02$, by discarding $5\%$ and $15\%$ of the acquired images, respectively. The significantly improved performance was observed for both tested tasks, compared with the respective $0.90\pm0.01$ and $0.82\pm0.02$ from networks without considering task amenability. This enables image quality feedback during real-time ultrasound acquisition among many other medical imaging applications.

Morphological Change Forecasting for Prostate Glands using Feature-based Registration and Kernel Density Extrapolation

Organ morphology is a key indicator for prostate disease diagnosis and prognosis. For instance, In longitudinal study of prostate cancer patients under active surveillance, the volume, boundary smoothness and their changes are closely monitored on time-series MR image data. In this paper, we describe a new framework for forecasting prostate morphological changes, as the ability to detect such changes earlier than what is currently possible may enable timely treatment or avoiding unnecessary confirmatory biopsies. In this work, an efficient feature-based MR image registration is first developed to align delineated prostate gland capsules to quantify the morphological changes using the inferred dense displacement fields (DDFs). We then propose to use kernel density estimation (KDE) of the probability density of the DDF-represented \textit{future morphology changes}, between current and future time points, before the future data become available. The KDE utilises a novel distance function that takes into account morphology, stage-of-progression and duration-of-change, which are considered factors in such subject-specific forecasting. We validate the proposed approach on image masks unseen to registration network training, without using any data acquired at the future target time points. The experiment results are presented on a longitudinal data set with 331 images from 73 patients, yielding an average Dice score of 0.865 on a holdout set, between the ground-truth and the image masks warped by the KDE-predicted-DDFs.

DeepReg: a deep learning toolkit for medical image registration

DeepReg (https://github.com/DeepRegNet/DeepReg) is a community-supported open-source toolkit for research and education in medical image registration using deep learning.

Assisted Probe Positioning for Ultrasound Guided Radiotherapy Using Image Sequence Classification

Effective transperineal ultrasound image guidance in prostate external beam radiotherapy requires consistent alignment between probe and prostate at each session during patient set-up. Probe placement and ultrasound image inter-pretation are manual tasks contingent upon operator skill, leading to interoperator uncertainties that degrade radiotherapy precision. We demonstrate a method for ensuring accurate probe placement through joint classification of images and probe position data. Using a multi-input multi-task algorithm, spatial coordinate data from an optically tracked ultrasound probe is combined with an image clas-sifier using a recurrent neural network to generate two sets of predictions in real-time. The first set identifies relevant prostate anatomy visible in the field of view using the classes: outside prostate, prostate periphery, prostate centre. The second set recommends a probe angular adjustment to achieve alignment between the probe and prostate centre with the classes: move left, move right, stop. The algo-rithm was trained and tested on 9,743 clinical images from 61 treatment sessions across 32 patients. We evaluated classification accuracy against class labels de-rived from three experienced observers at 2/3 and 3/3 agreement thresholds. For images with unanimous consensus between observers, anatomical classification accuracy was 97.2% and probe adjustment accuracy was 94.9%. The algorithm identified optimal probe alignment within a mean (standard deviation) range of 3.7$^{\circ}$ (1.2$^{\circ}$) from angle labels with full observer consensus, comparable to the 2.8$^{\circ}$ (2.6$^{\circ}$) mean interobserver range. We propose such an algorithm could assist ra-diotherapy practitioners with limited experience of ultrasound image interpreta-tion by providing effective real-time feedback during patient set-up.

Longitudinal Image Registration with Temporal-order and Subject-specificity Discrimination

Morphological analysis of longitudinal MR images plays a key role in monitoring disease progression for prostate cancer patients, who are placed under an active surveillance program. In this paper, we describe a learning-based image registration algorithm to quantify changes on regions of interest between a pair of images from the same patient, acquired at two different time points. Combining intensity-based similarity and gland segmentation as weak supervision, the population-data-trained registration networks significantly lowered the target registration errors (TREs) on holdout patient data, compared with those before registration and those from an iterative registration algorithm. Furthermore, this work provides a quantitative analysis on several longitudinal-data-sampling strategies and, in turn, we propose a novel regularisation method based on maximum mean discrepancy, between differently-sampled training image pairs. Based on 216 3D MR images from 86 patients, we report a mean TRE of 5.6 mm and show statistically significant differences between the different training data sampling strategies.

Image quality assessment for closed-loop computer-assisted lung ultrasound

We describe a novel, two-stage computer assistance system for lung anomaly detection using ultrasound imaging in the intensive care setting to improve operator performance and patient stratification during coronavirus pandemics. The proposed system consists of two deep-learning-based models. A quality assessment module automates predictions of image quality, and a diagnosis assistance module determines the likelihood-of-anomaly in ultrasound images of sufficient quality. Our two-stage strategy uses a novelty detection algorithm to address the lack of control cases available for training a quality assessment classifier. The diagnosis assistance module can then be trained with data that are deemed of sufficient quality, guaranteed by the closed-loop feedback mechanism from the quality assessment module. Integrating the two modules yields accurate, fast, and practical acquisition guidance and diagnostic assistance for patients with suspected respiratory conditions at the point-of-care. Using more than 25,000 ultrasound images from 37 COVID-19-positive patients scanned at two hospitals, plus 12 control cases, this study demonstrates the feasibility of using the proposed machine learning approach. We report an accuracy of 86% when classifying between sufficient and insufficient quality images by the quality assessment module. For data of sufficient quality, the mean classification accuracy in detecting COVID-19-positive cases was 95% on five holdout test data sets, unseen during the training of any networks within the proposed system.

Multimodality Biomedical Image Registration using Free Point Transformer Networks

We describe a point-set registration algorithm based on a novel free point transformer (FPT) network, designed for points extracted from multimodal biomedical images for registration tasks, such as those frequently encountered in ultrasound-guided interventional procedures. FPT is constructed with a global feature extractor which accepts unordered source and target point-sets of variable size. The extracted features are conditioned by a shared multilayer perceptron point transformer module to predict a displacement vector for each source point, transforming it into the target space. The point transformer module assumes no vicinity or smoothness in predicting spatial transformation and, together with the global feature extractor, is trained in a data-driven fashion with an unsupervised loss function. In a multimodal registration task using prostate MR and sparsely acquired ultrasound images, FPT yields comparable or improved results over other rigid and non-rigid registration methods. This demonstrates the versatility of FPT to learn registration directly from real, clinical training data and to generalize to a challenging task, such as the interventional application presented.