Deep learning algorithms for detection of critical findings in head CT scans: a retrospective study.
By Sasank Chilamkurthy, Rohit Ghosh, Swetha Tanamala, Mustafa Biviji, Norbert G. Campeau, V. Venugopal, V. Mahajan, Pooja Rao, Prashant Warier. The Lancet, 2018
837 citations recorded by the source corpus.
BACKGROUND Non-contrast head CT scan is the current standard for initial imaging of patients with head trauma or stroke symptoms. We aimed to develop and validate a set of deep learning algorithms for automated detection of the following key findings from these scans: intracranial haemorrhage and its types (ie, intraparenchymal, intraventricular, subdural, extradural, and subarachnoid); calvarial fractures; midline shift; and mass effect. METHODS We retrospectively collected a dataset containing 313 318 head CT scans together with their clinical reports from around 20 centres in India between Jan 1, 2011, and June 1, 2017. A randomly selected part of this dataset (Qure25k dataset) was used for validation and the rest was used to develop algorithms. An additional validation dataset (CQ500 dataset) was collected in two batches from centres that were different from those used for the development and Qure25k datasets. We excluded postoperative scans and scans of patients younger than 7 years. The original clinical radiology report and consensus of three independent radiologists were considered as gold standard for the Qure25k and CQ500 datasets, respectively. Areas under the receiver operating characteristic curves (AUCs) were primarily used to assess the algorithms. FINDINGS The Qure25k dataset contained 21 095 scans (mean age 43 years; 9030 [43%] female patients), and the CQ500 dataset consisted of 214 scans in the first batch (mean age 43 years; 94 [44%] female patients) and 277 scans in the second batch (mean age 52 years; 84 [30%] female patients). On the Qure25k dataset, the algorithms achieved an AUC of 0·92 (95% CI 0·91-0·93) for detecting intracranial haemorrhage (0·90 [0·89-0·91] for intraparenchymal, 0·96 [0·94-0·97] for intraventricular, 0·92 [0·90-0·93] for subdural, 0·93 [0·91-0·95] for extradural, and 0·90 [0·89-0·92] for subarachnoid). On the CQ500 dataset, AUC was 0·94 (0·92-0·97) for intracranial haemorrhage (0·95 [0·93-0·98], 0·93 [0·87-1·00], 0·95 [0·91-0·99], 0·97 [0·91-1·00], and 0·96 [0·92-0·99], respectively). AUCs on the Qure25k dataset were 0·92 (0·91-0·94) for calvarial fractures, 0·93 (0·91-0·94) for midline shift, and 0·86 (0·85-0·87) for mass effect, while AUCs on the CQ500 dataset were 0·96 (0·92-1·00), 0·97 (0·94-1·00), and 0·92 (0·89-0·95), respectively. INTERPRETATION Our results show that deep learning algorithms can accurately identify head CT scan abnormalities requiring urgent attention, opening up the possibility to use these algorithms to automate the triage process. FUNDING Qure.ai.
Subfields: Epidemiology, Neurology
View Paper: https://www.semanticscholar.org/paper/a1e37e17015efc88bf72e7e858523b9dd65ed35c
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Epidemiology
In Neurology
Citations in the corpus
- Key challenges for delivering clinical impact with artificial intelligence (2019, background): “For simpler tasks, including medical image classification, this problem may be less crucial and overcome by the curation of large, heterogenous, multi-centre datasets [14].”
- Clinically Applicable AI System for Accurate Diagnosis, Quantitative Measurements, and Prognosis of COVID-19 Pneumonia Using Computed Tomography (2020, background): “Another study showed that deep learning algorithms helped to identify head CT abnormalities that can assist the clinical triage process (Chilamkurthy et al., 2018).”
- Unsupervised Brain Abnormality Detection Using High Fidelity Image Reconstruction Networks (2020, background): “More-over, although supervised trained models can recognize learned entities well, they sometimes overlook abnormalities that are apparent to the human eye if no specific definition is provided in advance (Chilamkurthy et al., 2018).”
- An explainable deep-learning algorithm for the detection of acute intracranial haemorrhage from small datasets (2018, methodology): “● Different interpolation degrees were used for classification (3, 5, 7, 9).”
- Making the Most of Text Semantics to Improve Biomedical Vision-Language Processing (2022, background): “Such data is collected daily in routine clinical practice, and common examples are X-ray images [18, 32, 75] or computed tomography scans [9, 18, 20, 71] paired with reports written by qualified medical experts.”
- A Weakly-Supervised Framework for COVID-19 Classification and Lesion Localization From Chest CT (2020, methodology): “Firstly, in the current emergency situation, the number of enrolled patients is relatively smaller compared with that used in previous studies [17], [34]; and patients enrolled in our study”
- Hidden stratification causes clinically meaningful failures in machine learning for medical imaging (2019, background): “Chilamkurthy et al. [9] present a subset analysis for different diagnostic categories of intracranial hemorrhage (e.g. sub-dural vs. subarachnoid) when designing a deep learning model for abnormality detection on head CT, but do not analyze differences in performance related to bleed size,…”