Sort:
Open Access Original Research Issue
Performance of ChatGPT and Radiology Residents on Ultrasonography Board-Style Questions
Advanced Ultrasound in Diagnosis and Therapy 2024, 8(4): 250-254
Published: 30 December 2024
Abstract PDF (2.1 MB) Collect
Downloads:17
Objective

This study aims to assess the performance of the Chat Generative Pre-Trained Transformer (ChatGPT), specifically versions GPT-3.5 and GPT-4, on ultrasonography board-style questions, and subsequently compare it with the performance of third-year radiology residents on the identical set of questions.

Methods

The study, conducted from May 19 to May 30, 2023, utilized a selection of 134 multiple-choice questions sourced from a commercial question bank for American Registry for Diagnostic Medical Sonography (ARDMS) examinations and imported into the ChatGPT model (encompassing GPT-3.5 and GPT-4 versions). ChatGPT’s responses were evaluated overall, by topic, and by GPT version. An identical question set was assigned to three third-year radiology residents, enabling a direct comparison of performances with ChatGPT.

Results

GPT-4 correctly responded to 82.1% of questions (110 of 134), significantly surpassing the performance of GPT-3.5 (P = 0.003), which correctly answered 66.4% of questions (89 of 134). Although GPT-3.5’s performance was statistically indistinguishable from the average performance of the radiology residents (66.7%, 89.3 of 134) (P = 0.969), there was a notable difference in the accuracy in question-answering accuracy between GPT-4 and the residents (P = 0.004).

Conclusions

ChatGPT demonstrated significant competency in responding to ultrasonography board-style questions, with the GPT-4 version markedly surpassing both its predecessor GPT-3.5 and the radiology residents.

Open Access Review Article Issue
Deep Learning in Ultrasound Localization Microscopy
Advanced Ultrasound in Diagnosis and Therapy 2024, 8(3): 86-92
Published: 30 September 2024
Abstract PDF (1.4 MB) Collect
Downloads:141

Ultrasound imaging holds a significant position in medical diagnostics due to its non-invasive and real-time capabilities. However, traditional ultrasound is constrained by the diffraction limit, making it challenging to capture fine blood vessels. Ultrasound localization microscopy (ULM) overcomes this limitation by achieving super-resolution imaging through tracking the trajectories of microbubbles (MBs) within microvasculature. This review summarizes the applications of deep learning (DL) techniques in ULM post-processing algorithms, including key steps such as beamforming, clutter filtering and denoising, localization, and tracking. Although DL shows great potential in improving ULM imaging quality and efficiency, current research mainly focuses on imaging algorithmic improvements rather than in-depth image analysis. In the future, with the accumulation of ULM image data, the powerful feature extraction capability of DL is expected to further advance ULM applications in disease prediction and diagnosis.

Open Access Original Research Issue
Using S-Detect to Improve Breast Ultrasound: The Different Combined Strategies Based on Radiologist Experience
Advanced Ultrasound in Diagnosis and Therapy 2022, 6(4): 180-187
Published: 01 December 2022
Abstract PDF (1.3 MB) Collect
Downloads:4
Objective

To investigate the best combined method of S-Detect, a computer-aided diagnosis (CAD) system, with breast ultrasound (US) according to radiologists'experience.

Methods

From March 2019 to June 2019, 259 breast masses in 255 women were included in this study.Ultrasonographic images of the target masses were prospectively analyzed by radiologists and CAD.Three combined methods, including method 1[selective downgrading combination for Breast Imaging Reporting and Data System (BI-RADS)4a lesions], method 2(selective upgrading combination for BI-RADS 3 lesions) and method 3(selective upgrading or downgrading combination for BI-RADS 3 or 4a lesions), were applied to interpret the CAD results.The sensitivity, specificity, the area under the receiver operating characteristic curve (AUC) of experienced or inexperienced radiologists before and after adding CAD results were compared using the histopathological results as a reference standard.

Results

In identifying breast malignancy, the AUC for CAD was similar to that of experienced radiologists (P=0.410), but higher than that of inexperienced radiologists (P=0.003).When combining CAD with experienced radiologists based on method 1 and combining CAD results with inexperienced radiologists based on method 3, the AUCs were significantly improved (P=0.024 and 0.003, respectively) compared to US alone, with significantly increased specificity (P<0.001 for both) and no significantly decreased sensitivity (P>0.05 for both).

Conclusion

The combination of CAD system and conventional ultrasound can improve ultrasound diagnostic performance in determining breast malignancy.The method 1 and method 3 combinations are respectively recommended for experienced and inexperienced radiologists when CAD is combined with conventional breast ultrasound.

Open Access Review Article Issue
Ultrasound Image Generation and Modality Conversion Based on Deep Learning
Advanced Ultrasound in Diagnosis and Therapy 2023, 7(2): 136-139
Published: 30 June 2023
Abstract PDF (11.1 MB) Collect
Downloads:41

Artificial intelligent (AI) based on deep learning has been used in medical imaging analysis for years. Improvements have been made in the diagnosis of various diseases with the help of deep learning. Multimodal medical imaging combines two or more imaging modalities, providing comprehensive diagnostic information of the diseases. However, some modality problems always exist in clinical practice. Recently, AI-based deep learning technologies have realized the modality conversion. Investigations on modality conversion have gradually been reported in order to acquire multimodal information. MRI images could be generated from CT images while ultrasound elastography could be generated from B mode ultrasonography. Continuous researches and development of new technologies around deep learning are still under investigation and provide huge clinical potentials in the future. The purpose of this review is to summarize an overview of the current applications and prospects of deep learning-based modality conversion of medical imaging.

Open Access Technical Development Issue
Distributed Cloud-based Ultrasound Platform: Innovative Pathway to Develop Ultrasound Imaging System
Advanced Ultrasound in Diagnosis and Therapy 2022, 6(1): 33-37
Published: 30 March 2022
Abstract PDF (2.9 MB) Collect
Downloads:7

The current medical ultrasound imaging device is mainly divided into console ultrasound device and portable ultrasound device. In this article, a new concept along with an innovative pathway to develop ultrasound imaging devices, namely distributed cloud-based ultrasound system (DCUS), was proposed. In DCUS, the ultrasound probes from multiple terminals are used to complete the transmission and reception as well as analog-to-digital conversion of ultrasonic signals, and upload the original radio frequency (RF) signals or in-phase and quadrature (IQ) signals to the cloud server through ultra-bandwidth high-speed communication technology, while the centralized cloud server platform finishes processing of ultrasonic signals and transmits and distributes ultrasound imaging to each related terminal in real time. Various artificial intelligence (AI) algorithms can also be deployed on the cloud-based platform to achieve AI-powered imaging optimization, protocol standardization, and assisted diagnosis. Thus, by utilizing new cloud-based platform and super-high transmission technology and combining the advantages of console ultrasound and portable ultrasound systems with flexibility, high imaging quality and intelligent features, DCUS could become whole new ultrasound system for medical imaging applications in foreseeable future.

Open Access Original Research Issue
Characterization of Breast Lesions: Comparison between Three-dimensional Ultrasound and Automated Volume Breast Ultrasound
Advanced Ultrasound in Diagnosis and Therapy 2021, 5(3): 204-211
Published: 30 September 2021
Abstract PDF (10.4 MB) Collect
Downloads:4
Objective

This study aimed to compare the diagnostic performance of three-dimensional ultrasound (3D-US) and automated breast volume scanner (ABVS) for the characterization of benign and malignant breast lesions.

Methods

Ninety patients who underwent surgery and preoperative conventional ultrasound (US), 3D-US, and ABVS examinations were enrolled in this study. The image quality and adjacent structures of the lesions in the coronal plane were compared. The combination of US, 3D-US, and ABVS for retraction phenomenon of the lesion was compared and the diagnostic performance of each combination was analyzed.

Results

ABVS displayed better image quality and adjacent structures than 3D-US (P < 0.001). The area under the curve (AUC) was 0.913, 0.842, and 0.871 for US, 3D-US, and ABVS, respectively. The AUC of the retraction phenomenon of the lesion was 0.732 and 0.810 for 3D-US and ABVS, respectively. When they were combined, US+ABVS showed the highest AUC of 0.924. No significant difference of diagnostic performances was found among conventional US, US+3D-US, and US+ABVS (P > 0.05).

Conclusions

Compared with 3D-US, ABVS seems to be superior in showing the retraction phenomenon of breast lesions and in the characterization of breast lesions alone or in combination with conventional US. Although no significant difference was observed between them, both ABVS and 3D-US provided valuable information in the coronal plane and improved our confidence level in breast lesion characterization, especially when combined with the conventional US.

Total 6