RESUMEN

Lung cancer is by far the leading cause of cancer death in the US. Recent studies have demonstrated the effectiveness of screening using low dose CT (LDCT) in reducing lung cancer related mortality. While lung nodules are detected with a high rate of sensitivity, this exam has a low specificity rate and it is still difficult to separate benign and malignant lesions. The ISBI 2018 Lung Nodule Malignancy Prediction Challenge, developed by a team from the Quantitative Imaging Network of the National Cancer Institute, was focused on the prediction of lung nodule malignancy from two sequential LDCT screening exams using automated (non-manual) algorithms. We curated a cohort of 100 subjects who participated in the National Lung Screening Trial and had established pathological diagnoses. Data from 30 subjects were randomly selected for training and the remaining was used for testing. Participants were evaluated based on the area under the receiver operating characteristic curve (AUC) of nodule-wise malignancy scores generated by their algorithms on the test set. The challenge had 17 participants, with 11 teams submitting reports with method description, mandated by the challenge rules. Participants used quantitative methods, resulting in a reporting test AUC ranging from 0.698 to 0.913. The top five contestants used deep learning approaches, reporting an AUC between 0.87 - 0.91. The team's predictor did not achieve significant differences from each other nor from a volume change estimate (p =.05 with Bonferroni-Holm's correction).

Asunto(s)

RESUMEN

Deep neural networks usually require large labeled datasets to construct accurate models; however, in many real-world scenarios, such as medical image segmentation, labelling data is a time-consuming and costly human (expert) intelligent task. Semi-supervised methods leverage this issue by making use of a small labeled dataset and a larger set of unlabeled data. In this article, we present a flexible framework for semi-supervised learning that combines the power of supervised methods that learn feature representations using state-of-the-art deep convolutional neural networks with the deep embedded clustering algorithm that assigns data points to clusters based on their probability distributions and feature representations learned by the networks. Our proposed semi-supervised learning algorithm based on deep embedded clustering (SSLDEC) learns feature representations via iterations by alternatively using labeled and unlabeled data points and computing target distributions from predictions. During this iterative procedure the algorithm uses labeled samples to keep the model consistent and tuned with labeling, as it simultaneously learns to improve feature representation and predictions. SSLDEC requires few hyper-parameters and thus does not need large labeled validation sets, which addresses one of the main limitations of many semi-supervised learning algorithms. It is also flexible and can be used with many state-of-the-art deep neural network configurations for image classification and segmentation tasks. To this end, we implemented and tested our approach on benchmark image classification tasks as well as in a challenging medical image segmentation scenario. In benchmark classification tasks, SSLDEC outperformed several state-of-the-art semi-supervised learning methods, achieving 0.46% error on MNIST with 1000 labeled points, and 4.43% error on SVHN with 500 labeled points. In the iso-intense infant brain MRI tissue segmentation task, we implemented SSLDEC on a 3D densely connected fully convolutional neural network where we achieved significant improvement over supervised-only training as well as a semi-supervised method based on pseudo-labelling. Our results show that SSLDEC can be effectively used to reduce the need for costly expert annotations, enhancing applications such as automatic medical image segmentation.