Automated detection of pulmonary embolism from CT-angiograms using deep learning

Abstract

Background The aim of this study was to develop and evaluate a deep neural network model in the automated detection of pulmonary embolism (PE) from computed tomography pulmonary angiograms (CTPAs) using only weakly labelled training data. Methods We developed a deep neural network model consisting of two parts: a convolutional neural network architecture called InceptionResNet V2 and a long-short term memory network to process whole CTPA stacks as sequences of slices. Two versions of the model were created using either chest X-rays (Model A) or natural images (Model B) as pre-training data. We retrospectively collected 600 CTPAs to use in training and validation and 200 CTPAs to use in testing. CTPAs were annotated only with binary labels on both stack- and slice-based levels. Performance of the models was evaluated with ROC and precision–recall curves, specificity, sensitivity, accuracy, as well as positive and negative predictive values. Results Both models performed well on both stack- and slice-based levels. On the stack-based level, Model A reached specificity and sensitivity of 93.5% and 86.6%, respectively, outperforming Model B slightly (specificity 90.7% and sensitivity 83.5%). However, the difference between their ROC AUC scores was not statistically significant (0.94 vs 0.91, p = 0.07). Conclusions We show that a deep learning model trained with a relatively small, weakly annotated dataset can achieve excellent performance results in detecting PE from CTPAs.

Convolutional architectures with LSTM and TCN to embolism classification: exploring dependency between data

“…Our approach is concerned with finding the best results within the conditions using LSTM and TCN, not seeking to optimize the classification of pulmonary embolism. Huhtanen et al (2022) presented the model precision at a value of 60% -70% using the confidence interval. However, the data used in the model were balanced, unlike the data presented in this project and the real scenario.…”

“…This work aims to compare the two main approaches in the classification of pulmonary embolism in CTPA ( [Huhtanen et al 2022], [Ma et al 2022]), analyzing these techniques from the field of anomaly detection. Our focus was to observe convolutional architectures with architectures that use temporal dependence as input.…”

“…The machine used to run the experiments had an Intel(R) Core(TM) i7-8700K processor with 64 GB of RAM and two Nvidia GTX 1080 TI video cards. The hyperparameters used for each model can be seen in Table 1, the hyperparameters selected were based on previous works [Huhtanen et al 2022, Ma et al 2022] and observing the loss function from the model. The models were created using the PyTorch library [MetaAI 2022], and the language for writing the code was Python.…”

“…For example, Ma et al (2022) proposed a deep learning multitask learning method using CNN and the Temporal Convolutional Network (TCN) to predict disease and other classifications, which resulted in good performance using this dataset. Another example is a deep learning experiment using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM), but using a private dataset [Huhtanen et al 2022]. According to these two works, we proposed comparing different approaches to explore dependency between the images along the exam using Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN).…”

Auto-detection of motion artifacts on CT pulmonary angiograms with a physician-trained AI algorithm

“…An automated method of detecting substantial motion artifacts at the time of scanning can help re-scan patients during the same imaging session with better coaching or faster scanning techniques. Recent advances in machine and deep learning (DL) have led to the creation of several AI algorithms in medical imaging including in the areas of image reconstruction, triaging, quality control and pathology detection [9][10][11][12][13][14]. Therefore, we trained and tested an artificial intelligence (AI) model to identify substantial motion artifacts on CTPA that have a negative impact on diagnostic interpretation.…”