Hyperparameter Optimization for COVID-19 Pneumonia Diagnosis Based on Chest CT

Abstract

Convolutional Neural Networks (CNNs) have been successfully applied in the medical diagnosis of different types of diseases. However, selecting the architecture and the best set of hyperparameters among the possible combinations can be a significant challenge. The purpose of this work is to investigate the use of the Hyperband optimization algorithm in the process of optimizing a CNN applied to the diagnosis of SARS-Cov2 disease (COVID-19). The test was performed with the Optuna framework, and the optimization process aimed to optimize four hyperparameters: (1) backbone architecture, (2) the number of inception modules, (3) the number of neurons in the fully connected layers, and (4) the learning rate. CNNs were trained on 2175 computed tomography (CT) images. The CNN that was proposed by the optimization process was a VGG16 with five inception modules, 128 neurons in the two fully connected layers, and a learning rate of 0.0027. The proposed method achieved a sensitivity, precision, and accuracy of 97%, 82%, and 88%, outperforming the sensitivity of the Real-Time Polymerase Chain Reaction (RT-PCR) tests (53–88%) and the accuracy of the diagnosis performed by human experts (72%).

Pulmonary COVID-19: Learning Spatiotemporal Features Combining CNN and LSTM Networks for Lung Ultrasound Video Classification

“…Transfer learning with models pre-trained on ImageNet provided comparable results to models pre-trained on LUS images, suggesting that ImageNet can be used in cases where there is limited data for training [ 99 ]. We also show that the use of transfer learning techniques and HPO can facilitate the creation of rapid prototypes for diagnosing diseases, as seen in other studies [ 21 ].…”

“…Studies based on patients affected by COVID-19 indicate a high prevalence of respiratory symptoms that require a comprehensive evaluation [ 18 , 19 ]. The standard method for diagnosing the disease involves a clinical examination, performing pulmonary auscultation of the patient using a stethoscope, and then additional imaging examinations such as X-ray (XR) and computed tomography (CT) [ 20 , 21 ]. The laboratory test known as reverse transcription-polymerase chain reaction (RT-PCR) is considered the gold standard [ 22 ] being used to diagnosis the disease.…”

“…Moreover, due to the portability and low cost of US equipment, such devices enable rapid prototyping of systems that make intensive use of AI and computer vision (CV) techniques to aid in the diagnosis of COVID-19 [ 39 ]. However, despite the advantages presented in the use of LUS images for the diagnosis of pulmonary pathology and COVID-19, few studies are exploring Deep Learning (DL) techniques compared to the number of studies performed with other medical imaging techniques such as XR and CT [ 21 , 40 , 41 , 42 , 43 ].…”

“…These techniques are predominantly applied to other types of medical imaging (XR and CT) [ 21 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 ], while studies that consider US imaging in their medical diagnostic models are less frequent. One of the possible causes for it is the lag of public databases available for scientific research, especially concerning LUS videos containing proved cases of COVID-19, according to a survey conducted by [ 60 ].…”

Per-COVID-19: A Benchmark Dataset for COVID-19 Percentage Estimation from CT-Scans

“…The state-of-the-Art methods using CT scans can be classified into two main tasks: COVID-19 recognition [5,6,[14][15][16] and COVID-19 segmentation [7,8,17,18]. In [19], Zheng, C. et al proposed the DeCoVNet approach, which is based on 3D deep convolutional neural Network to Detect COVID-19 (DeCoVNet) from CT volumes.…”