DeepCADe: A Deep Learning Architecture for the Detection of Lung Nodules in CT Scans
Bibliographic record
Abstract
Early detection of lung nodules in thoracic Computed Tomography (CT) scans is of great importance for the successful diagnosis and treatment of lung cancer. Due to improvements in screening technologies, and an increased demand for their use, radiologists are required to analyze an ever increasing amount of image data, which can affect the quality of their diagnoses. Computer-Aided Detection (CADe) systems are designed to assist radiologists in this endeavor. In this thesis, we present DeepCADe, a novel CADe system for the detection of lung nodules in thoracic CT scans which produces improved results compared to the state-of-the-art in this field of research. CT scans are grayscale images, so the terms scans and images are used interchangeably in this work. DeepCADe was trained with the publicly available Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) database, which contains 1018 thoracic CT scans with nodules of different shape and size, and is built on a Deep Convolutional Neural Network (DCNN), which is trained using the backpropagation algorithm to extract volumetric features from the input data and detect lung nodules in sub-volumes of CT images. Considering only lung nodules that have been annotated by at least three radiologists, DeepCADe achieves a 2.1% improvement in sensitivity (true positive rate) over the best result in the current published scientific literature, assuming an equal number of false positives (FPs) per scan. More specifically, it achieves a sensitivity of 89.6% with 4 FPs per scan, or a sensitivity of 92.8% with 10 FPs per scan. Furthermore, DeepCADe is validated on a larger number of lung nodules compared to other studies (Table 5.2). This increases the variation in the appearance of nodules and therefore makes their detection by a CADe system more challenging. We study the application of Deep Convolutional Neural Networks (DCNNs) for the detection of lung nodules in thoracic CT scans. We explore some of the meta parameters that affect the performance of such models, which include: 1. the network architecture, i.e. its structure in terms of convolution layers, fully-connected layers, pooling layers, and activation functions, 2. the receptive field of the network, which defines the dimensions of its input, i.e. how much of the CT scan is processed by the network in a single forward pass, 3. a threshold value, which affects the sliding window algorithm with which the network is used to detect nodules in complete CT scans, and 4. the agreement level, which is used to interpret the independent nodule annotations of four experienced radiologists. Finally, we visualize the shape and location of annotated lung nodules and compare them to the output of DeepCADe. This demonstrates the compactness and flexibility in shape of the nodule predictions made by our proposed CADe system. In addition to the 5-fold cross validation results presented in this thesis, these visual results support the applicability of our proposed CADe system in real-world medical practice.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".