Search

In recent years, artificial intelligence, especially object detection-based deep learning in computer vision, has made significant advancements, driven by the development of computing power and the widespread use of graphic processor units. Object detection-based deep learning techniques have been applied in various fields, including the medical imaging domain, where remarkable achievements have been reported in disease detection. However, the application of deep learning does not always guarantee satisfactory performance, and researchers have been employing trial-and-error to identify the factors contributing to performance degradation and enhance their models. Moreover, due to the black-box problem, the intermediate processes of a deep learning network cannot be comprehended by humans; as a result, identifying problems in a deep learning model that exhibits poor performance can be challenging. This article highlights potential issues that may cause performance degradation at each deep learning step in the medical imaging domain and discusses factors that must be considered to improve the performance of deep learning models. Researchers who wish to begin deep learning research can reduce the required amount of trial-and-error by understanding the issues discussed in this study., Competing Interests: The authors have no potential conflicts of interest to disclose., (Copyright © 2023 The Korean Society of Radiology.)

Brain metastases (BM) are the most common intracranial tumors, and their prevalence is increasing. High-resolution black-blood (BB) imaging was used to complement the conventional contrast-enhanced 3D gradient-echo imaging to detect BM. In this study, we propose an efficient deep learning algorithm (DLA) for BM detection in BB imaging with contrast enhancement scans, and assess the efficacy of an automatic detection algorithm for BM. A total of 113 BM participants with 585 metastases were included in the training cohort for five-fold cross-validation. The You Only Look Once (YOLO) V2 network was trained with 3D BB sampling perfection with application-optimized contrasts using different flip angle evolution (SPACE) images to investigate the BM detection. For the observer performance, two board-certified radiologists and two second-year radiology residents detected the BM and recorded the reading time. For the training cohort, the overall performance of the five-fold cross-validation was 87.95%, 24.82%, 19.35%, 14.48, and 18.40 for sensitivity, precision, F1-Score, the false positive average for the BM dataset, and the false positive average for the normal individual dataset, respectively. For the comparison of reading time with and without DLA, the average reading time was reduced by 20.86% in the range of 15.22-25.77%. The proposed method has the potential to detect BM with a high sensitivity and has a limited number of false positives using BB imaging., (© 2022. The Author(s).)

Objective: Mammography is the initial examination to detect breast cancer symptoms, and quality control of mammography devices is crucial to maintain accurate diagnosis and to safeguard against degradation of performance. The objective of this study was to assist radiologists in mammography phantom image evaluation by developing and validating an interpretable deep learning model capable of objectively evaluating the quality of standard phantom images for mammography., Materials and Methods: A total of 2,208 mammography phantom images were collected for periodic accreditation of the scanner from 1,755 institutions. The dataset was randomly split into training (1,808 images) and testing (400 images) datasets with subgroups (76 images) for the multi-reader study. To develop an interpretable model that contains two deep learning networks in series, five processing steps were performed: mammography phantom detection, phantom object detection, post-processing, score evaluation, and a report with a comment about ambiguous results., Results: For phantom detection, the accuracy and mean intersection over union (mIOU) were 1.00 and 0.938 in the test dataset, respectively. During phantom object detection, a total of 6,369 out of 6,400 objects were detected as the correct object class, and the accuracy and mIOU were 0.995 and 0.813, respectively. The predicted score for each object showed a consensus of 97.40% excluding ambiguous points and 59.10% for ambiguous points of the groups., Conclusions: The interpretable deep learning model using large-scale data from multiple centers shows high performance and reasonable object scoring, successfully validating the reliability and feasibility of mammography phantom image quality management., (Copyright © 2022 Elsevier B.V. All rights reserved.)

Purpose: This study aimed to propose an effective end-to-end process in medical imaging using an independent task learning (ITL) algorithm and to evaluate its performance in maxillary sinusitis applications., Materials and Methods: For the internal dataset, 2122 Waters' view X-ray images, which included 1376 normal and 746 sinusitis images, were divided into training (n=1824) and test (n=298) datasets. For external validation, 700 images, including 379 normal and 321 sinusitis images, from three different institutions were evaluated. To develop the automatic diagnosis system algorithm, four processing steps were performed: 1) preprocessing for ITL, 2) facial patch detection, 3) maxillary sinusitis detection, and 4) a localization report with the sinusitis detector., Results: The accuracy of facial patch detection, which was the first step in the end-to-end algorithm, was 100%, 100%, 99.5%, and 97.5% for the internal set and external validation sets #1, #2, and #3, respectively. The accuracy and area under the receiver operating characteristic curve (AUC) of maxillary sinusitis detection were 88.93% (0.89), 91.67% (0.90), 90.45% (0.86), and 85.13% (0.85) for the internal set and external validation sets #1, #2, and #3, respectively. The accuracy and AUC of the fully automatic sinusitis diagnosis system, including site localization, were 79.87% (0.80), 84.67% (0.82), 83.92% (0.82), and 73.85% (0.74) for the internal set and external validation sets #1, #2, and #3, respectively., Conclusion: ITL application for maxillary sinusitis showed reasonable performance in internal and external validation tests, compared with applications used in previous studies., Competing Interests: The authors have no potential conflicts of interest to disclose., (© Copyright: Yonsei University College of Medicine 2021.)

Purpose: Cerebral microbleeds (CMBs) are considered essential indicators for the diagnosis of cerebrovascular disease and cognitive disorders. Traditionally, CMBs are manually interpreted based on criteria including the shape, diameter, and signal characteristics after an MR examination, such as susceptibility-weighted imaging or gradient echo imaging (GRE). In this paper, an efficient method for CMB detection in GRE scans is presented., Materials and Methods: The proposed framework consists of the following phases: (1) pre-processing (skull extraction), (2) the first training with the ground truth labeled using CMB, (3) the second training with the ground truth labeled with CMB mimicking the same subjects, and (4) post-processing (cerebrospinal fluid (CSF) filtering). The proposed technique was validated on a dataset of 1133 CBMs that consisted of 5284 images for training and 1737 images for testing. We applied a two-stage approach using a region-based CNN method based on You Only Look Once (YOLO) to investigate a novel CMB detection technique., Results: The sensitivity, precision, F1-score and false positive per person (FP avg) were evaluated as 80.96, 60.98, 69.57 and 6.57, 59.69, 62.70, 61.16 and 4.5, 66.90, 79.75, 72.76 and 2.15 for YOLO with a single label, YOLO with double labels, and YOLO + CSF filtering, respectively, and YOLO + CSF filtering showed the highest precision performance, F1-score and lowest FP avg ., Conclusions: Using proposed framework, we developed an optimized CMB learning model with low false positives and a balanced performance in clinical practice., Competing Interests: Declaration of Competing Interest We declare that there are no conflicts of interest., (Copyright © 2021. Published by Elsevier Inc.)

The rapid development of deep learning in medical imaging has significantly enhanced the capabilities of artificial intelligence while simultaneously introducing challenges, including the need for vast amounts of training data and the labor-intensive tasks of labeling and segmentation. Generative adversarial networks (GANs) have emerged as a solution, offering synthetic image generation for data augmentation and streamlining medical image processing tasks through models such as cGAN, CycleGAN, and StyleGAN. These innovations not only improve the efficiency of image augmentation, reconstruction, and segmentation, but also pave the way for unsupervised anomaly detection, markedly reducing the reliance on labeled datasets. Our investigation into GANs in medical imaging addresses their varied architectures, the considerations for selecting appropriate GAN models, and the nuances of model training and performance evaluation. This paper aims to provide radiologists who are new to GAN technology with a thorough understanding, guiding them through the practical application and evaluation of GANs in brain imaging with two illustrative examples using CycleGAN and pixel2style2pixel (pSp)-combined StyleGAN. It offers a comprehensive exploration of the transformative potential of GANs in medical imaging research. Ultimately, this paper strives to equip radiologists with the knowledge to effectively utilize GANs, encouraging further research and application within the field. [ABSTRACT FROM AUTHOR]