Abstract
Due to the fast transmission speed and severe health damage, COVID-19 has attracted global attention. Early diagnosis and isolation are effective and imperative strategies for epidemic prevention and control. Most diagnostic methods for the COVID-19 is based on nucleic acid testing (NAT), which is expensive and time-consuming. To build an efficient and valid alternative of NAT, this article investigates the feasibility of employing computed tomography images of lungs as the diagnostic signals. Unlike normal lungs, parts of the lungs infected with the COVID-19 developed lesions, ground-glass opacity, and bronchiectasis became apparent. Through a public dataset, in this article, we propose an advanced residual learning diagnosis detection (RLDD) scheme for the COVID-19 technique, which is designed to distinguish positive COVID-19 cases from heterogeneous lung images. Besides the advantage of high diagnosis effectiveness, the designed residual-based COVID-19 detection network can efficiently extract the lung features through small COVID-19 samples, which removes the pretraining requirement on other medical datasets. In the test set, we achieve an accuracy of 91.33%, a precision of 91.30%, and a recall of 90%. For the batch of 150 samples, the assessment time is only 4.7 s. Therefore, RLDD can be integrated into the application programming interface and embedded into the medical instrument to improve the detection efficiency of COVID-19.
Introduction
Standard COVID-19 diagnosis relies heavily on RT-PCR (nucleic acid) testing, which is often slow, expensive, and requires close physical contact between medical staff and patients. While Computed Tomography (CT) scans of the lungs offer a faster alternative, manually interpreting these images is subjective and places a heavy burden on radiologists.
Researchers have developed the Residual Learning Diagnosis Detection (RLDD) system, a deep learning framework that identifies COVID-19 from lung CT scans with over 91% accuracy. By integrating this system into the Industrial Internet of Things (IIoT), the authors provide a path for rapid, non-contact, and automated screening that bypasses the long wait times of traditional lab tests.
Methods and System Architecture
The core of the system is the RLDD network, a 15-layer Convolutional Neural Network (CNN) built on the principle of Residual Learning. Instead of forcing every layer to learn a brand-new representation of the image, residual blocks allow the network to learn the "residual" (the difference) between the input and the output. This prevents the "vanishing gradient" problem common in deep networks and allows for faster convergence.
To handle inconsistent image data, the system uses a robust preprocessing pipeline. This includes pixel normalization to the [0, 1] range, resizing images to $224 \times 224$ pixels, and data augmentation through random cropping and horizontal flipping. This pipeline ensures the model is resilient to variations in scan quality and orientation.
Furthermore, the system is designed for IIoT integration. CT scanners act as IoT nodes that automatically upload images to a cloud or local server. The RLDD algorithm processes these images and sends results to a doctor’s terminal, enabling remote diagnosis and reducing infection risks for healthcare workers.
Dataset and Clinical Markers
The study utilized the public COVID-CT dataset, consisting of 349 positive and 347 negative CT samples. The dataset includes heterogeneous lung images showing various stages of infection. Clinical markers such as ground-glass opacity, lesions, and bronchiectasis are key features the network learns to identify.
The RLDD architecture is specifically designed to extract valid features from small sample sizes. This removes the common requirement in medical AI to pre-train on massive datasets like "ChestX-ray14," making it easier to deploy for emerging diseases where data may be scarce.
Key Results
The RLDD system demonstrated impressive performance on the test set:
- Accuracy: 91.33%
- Precision: 91.30%
- Recall: 90.00%
- Speed: The system processed a batch of 150 samples in just 4.7 seconds (approximately 0.03 seconds per image) using a mid-range GPU (NVIDIA GTX 950 M).
Compared to previous benchmarks on the same dataset, which hovered around 76% recall, this method represents a significant leap in reliability for identifying positive cases.
Conclusion and Future Work
While the results are promising, the authors acknowledge several areas for improvement. Future work involves collaborating with hospitals to build a more comprehensive, large-scale dataset to improve the model's robustness. Additionally, the researchers plan to introduce an "attention mechanism" to the CNN to help the algorithm focus specifically on the lung lobes and ignore irrelevant "noise" at the edges of the CT image. Advanced data cleaning processes will also be necessary for seamless integration into diverse clinical environments with varying image quality.
Summary updated