Tuberculosis (TB) remains a significant global health challenge, with pediatric cases posing a major concern. The World Health Organization (WHO) advocates for chest X-rays (CXRs) for TB screening. However, visual interpretation by radiologists can be subjective, time-consuming and prone to error, especially in pediatric TB. Artificial intelligence (AI)-driven computer-aided detection (CAD) tools, especially those utilizing deep learning, show promise in enhancing lung disease detection. However, challenges include data scarcity and lack of generalizability. In this context, we propose a novel self-supervised paradigm leveraging Vision Transformers (ViT) for improved TB detection in CXR, enabling zero-shot pediatric TB detection. We demonstrate improvements in TB detection performance ($\sim$12.7% and $\sim$13.4% top AUC/AUPR gains in adults and children, respectively) when conducting self-supervised pre-training when compared to fully-supervised (i.e., non pre-trained) ViT models, achieving top performances of 0.959 AUC and 0.962 AUPR in adult TB detection, and 0.697 AUC and 0.607 AUPR in zero-shot pediatric TB detection. As a result, this work demonstrates that self-supervised learning on adult CXRs effectively extends to challenging downstream tasks such as pediatric TB detection, where data are scarce.
Image harmonization is an important preprocessing strategy to address domain shifts arising from data acquired using different machines and scanning protocols in medical imaging. However, benchmarking the effectiveness of harmonization techniques has been a challenge due to the lack of widely available standardized datasets with ground truths. In this context, we propose three metrics: two intensity harmonization metrics and one anatomy preservation metric for medical images during harmonization, where no ground truths are required. Through extensive studies on a dataset with available harmonization ground truth, we demonstrate that our metrics are correlated with established image quality assessment metrics. We show how these novel metrics may be applied to real-world scenarios where no harmonization ground truth exists. Additionally, we provide insights into different interpretations of the metric values, shedding light on their significance in the context of the harmonization process. As a result of our findings, we advocate for the adoption of these quantitative harmonization metrics as a standard for benchmarking the performance of image harmonization techniques.
For machine learning-based prognosis and diagnosis of rare diseases, such as pediatric brain tumors, it is necessary to gather medical imaging data from multiple clinical sites that may use different devices and protocols. Deep learning-driven harmonization of radiologic images relies on generative adversarial networks (GANs). However, GANs notoriously generate pseudo structures that do not exist in the original training data, a phenomenon known as "hallucination". To prevent hallucination in medical imaging, such as magnetic resonance images (MRI) of the brain, we propose a one-shot learning method where we utilize neural style transfer for harmonization. At test time, the method uses one image from a clinical site to generate an image that matches the intensity scale of the collaborating sites. Our approach combines learning a feature extractor, neural style transfer, and adaptive instance normalization. We further propose a novel strategy to evaluate the effectiveness of image harmonization approaches with evaluation metrics that both measure image style harmonization and assess the preservation of anatomical structures. Experimental results demonstrate the effectiveness of our method in preserving patient anatomy while adjusting the image intensities to a new clinical site. Our general harmonization model can be used on unseen data from new sites, making it a valuable tool for real-world medical applications and clinical trials.
Chest X-rays (CXRs) are a widely used imaging modality for the diagnosis and prognosis of lung disease. The image analysis tasks vary. Examples include pathology detection and lung segmentation. There is a large body of work where machine learning algorithms are developed for specific tasks. A significant recent example is Coronavirus disease (covid-19) detection using CXR data. However, the traditional diagnostic tool design methods based on supervised learning are burdened by the need to provide training data annotation, which should be of good quality for better clinical outcomes. Here, we propose an alternative solution, a new self-supervised paradigm, where a general representation from CXRs is learned using a group-masked self-supervised framework. The pre-trained model is then fine-tuned for domain-specific tasks such as covid-19, pneumonia detection, and general health screening. We show that the same pre-training can be used for the lung segmentation task. Our proposed paradigm shows robust performance in multiple downstream tasks which demonstrates the success of the pre-training. Moreover, the performance of the pre-trained models on data with significant drift during test time proves the learning of a better generic representation. The methods are further validated by covid-19 detection in a unique small-scale pediatric data set. The performance gain in accuracy (~25\%) is significant when compared to a supervised transformer-based method. This adds credence to the strength and reliability of our proposed framework and pre-training strategy.
Disentangled representations can be useful in many downstream tasks, help to make deep learning models more interpretable, and allow for control over features of synthetically generated images that can be useful in training other models that require a large number of labelled or unlabelled data. Recently, flow-based generative models have been proposed to generate realistic images by directly modeling the data distribution with invertible functions. In this work, we propose a new flow-based generative model framework, named GLOWin, that is end-to-end invertible and able to learn disentangled representations. Feature disentanglement is achieved by factorizing the latent space into components such that each component learns the representation for one generative factor. Comprehensive experiments have been conducted to evaluate the proposed method on a public brain tumor MR dataset. Quantitative and qualitative results suggest that the proposed method is effective in disentangling the features from complex medical images.
High-quality labeled data is essential to successfully train supervised machine learning models. Although a large amount of unlabeled data is present in the medical domain, labeling poses a major challenge: medical professionals who can expertly label the data are a scarce and expensive resource. Making matters worse, voxel-wise delineation of data (e.g. for segmentation tasks) is tedious and suffers from high inter-rater variance, thus dramatically limiting available training data. We propose a recursive training strategy to perform the task of semantic segmentation given only very few training samples with pixel-level annotations. We expand on this small training set having cheaper image-level annotations using a recursive training strategy. We apply this technique on the segmentation of intracranial hemorrhage (ICH) in CT (computed tomography) scans of the brain, where typically few annotated data is available.
Semantic segmentation is an import task in the medical field to identify the exact extent and orientation of significant structures like organs and pathology. Deep neural networks can perform this task well by leveraging the information from a large well-labeled data-set. This paper aims to present a method that mitigates the necessity of an extensive well-labeled data-set. This method also addresses semi-supervision by enabling segmentation based on bounding box annotations, avoiding the need for full pixel-level annotations. The network presented consists of a single U-Net based unbranched architecture that generates a few-shot segmentation for an unseen human organ using just 4 example annotations of that specific organ. The network is trained by alternately minimizing the nearest neighbor loss for prototype learning and a weighted cross-entropy loss for segmentation learning to perform a fast 3D segmentation with a median score of 54.64%.