Radiological Interpretation Research Articles

Celiac Trunk with a Replaced Right Hepatic Artery: A Rare Anatomical Variant

The gastrointestinal tract and its accessory organs are supplied by the three main ventral branches of the abdominal aorta; namely, the celiac trunk, superior mesenteric artery, and inferior mesenteric artery. Variations of these vessels are related to the embryological development of the different parts of the gut; foregut, midgut, and hindgut. We report a case of a replaced right hepatic artery originating from the celiac trunk. Such variations are clinically important for the success of radiological image interpretation and surgical procedures of the region.

Machine learning applications in breast cancer prediction using mammography

Breast cancer is the second leading cause of cancer-related deaths among women. Early detection of lumps and subsequent risk assessment significantly improves prognosis. In screening mammography, radiologist interpretation of mammograms is prone to high error rates and requires extensive manual effort. To this end, several computer-aided diagnosis methods using machine learning have been proposed for automatic detection of breast cancer in mammography. In this paper, we provide a comprehensive review and analysis of these methods and discuss practical issues associated with their reproducibility. We aim to aid the readers in choosing the appropriate method to implement and we guide them towards this purpose. Moreover, an effort is made to re-implement a sample of the presented methods in order to highlight the importance of providing technical details associated with those methods. Advancing the domain of breast cancer pathology classification using machine learning involves the availability of public databases and development of innovative methods. Although there is significant progress in both areas, more transparency in the latter would boost the domain progress.

Comparative Analysis of Large Language Models and Spine Surgeons in Surgical Decision-Making and Radiological Assessment for Spine Pathologies

AimThis study aimed to investigate the accuracy of large language models (LLMs), specifically ChatGPT and Claude, in surgical decision-making and radiological assessment for spine pathologies compared to experienced spine surgeons. MethodsThe study employed a comparative analysis between the LLMs and a panel of attending spine surgeons. Five written clinical scenarios encompassing various spine pathologies were presented to the LLMs and surgeons, who provided recommended surgical treatment plans. Additionally, MRI images depicting spine pathologies were analyzed by the LLMs and surgeons to assess their radiological interpretation abilities. Spino-pelvic parameters were estimated from a scoliosis radiograph by the LLMs. ResultsQualitative content analysis revealed limitations in the LLMs' consideration of patient-specific factors and the breadth of treatment options. Both ChatGPT and Claude provided detailed descriptions of MRI findings but differed from the surgeons in terms of specific levels and severity of pathologies. The LLMs acknowledged the limitations of accurately measuring spino-pelvic parameters without specialized tools. The accuracy of surgical decision-making for the LLMs (20%) was lower than that of the attending surgeons (100%). Statistical analysis showed no significant differences in accuracy between the groups. ConclusionThe study highlights the potential of LLMs in assisting with radiological interpretation and surgical decision-making in spine surgery. However, the current limitations, such as the lack of consideration for patient-specific factors and inaccuracies in treatment recommendations, emphasize the need for further refinement and validation of these AI models. Continued collaboration between AI researchers and clinical experts is crucial to address these challenges and realize the full potential of AI in spine surgery.

Skeletal radiograph interpretation discrepancies in the emergency department setting: A retrospective chart review.

To determine the frequency and clinical consequences of discrepancies in skeletal radiograph interpretation between emergency and radiology doctors in an Australian ED. We reviewed the records of adult and paediatric patients assessed with skeletal radiography in an ED in Victoria, Australia over 3 months (January to March 2022). Epidemiological data, the interpretation of the radiograph by ED and radiology doctors, and clinical management of the patient were recorded to determine interpretation discrepancies and the consequences of these. There were 2359 unique skeletal radiographs in 1576 patient presentations during the study period. Of these, 140 (6%) had a discrepancy. Where a discrepancy existed, 47% of the ED interpretation reported a fracture and/or dislocation which was not present in the radiology interpretation (false positive), whereas the remaining (53%) were attributed to a missed fracture and/or dislocation (false negative). Thirty-five discrepancies (2%) required a change in patient management and were therefore clinically significant. The most commonly affected body region was the elbow, where 15% of radiographs were discrepant. Pathology was more often missed when multiple abnormalities were present on the same radiograph (odds ratio = 4.2, 95% confidence interval = 2.5-6.8). The rate of clinically significant discrepancies in the interpretation of skeletal radiographs by emergency medicine doctors is low. This data support using the ED interpretation of radiographs to guide initial management as safe practice.

Development of a deep learning method to identify acute ischaemic stroke lesions on brain CT

BackgroundCT is commonly used to image patients with ischaemic stroke but radiologist interpretation may be delayed. Machine learning techniques can provide rapid automated CT assessment but are usually developed from...

Quantification of Interstitial Lung Diseases, From the AJR Special Series on Quantitative Imaging.

High-resolution CT (HRCT) plays an important role in diagnosing and monitoring interstitial lung diseases (ILDs). Despite advances, predicting disease progression and treatment response remains challenging. HRCT enables noninvasive visualization and classification of patterns of lung injury and assessment of disease extent. Visual estimation of CT extent of fibrotic lung disease is an independent predictor of mortality and progression, but is subjective, with only modest interobserver agreement for radiologic interpretation of ILD. Machine learning-based textural analysis of fibrosis extent on baseline and serial HRCT scans shows robust correlations with physiologic measures and strong association with risk of disease progression or mortality across various fibrosing ILDs. In idiopathic pulmonary fibrosis, quantitative CT (QCT) assessment is associated with physiologic impairment and risk of progression and death, and increasing severity of fibrosis on longitudinal evaluation is associated with increased risk of progression and death. Similar results have been noted for fibrotic hypersensitivity pneumonitis and connective tissue disease. This review focuses on QCT techniques for ILDs. We describe the clinical need for quantification of lung disease and illustrate the role of conventional visual evaluation and of QCT approaches in defining disease severity, prognosis, and longitudinal progression, both in established disease and in preclinical interstitial abnormality.

Early Ischemic Stroke Assessment with ASPECTS: A Case Report Highlighting the Radiologist's Role in a Limited-Resource Setting

Background: Ischemic stroke is a leading cause of morbidity and mortality globally, particularly in resource-limited settings. Non-contrast computed tomography (NCCT) is often the primary imaging modality available in these settings, and the Alberta Stroke Program Early CT Score (ASPECTS) is a crucial tool for assessing early ischemic changes in NCCT. This case report highlights the importance of ASPECTS in guiding clinical decisions and prognostication in a resource-limited setting. Case presentation: A 79-year-old male presented to the emergency unit at Negara General Hospital, a rural facility in Bali, with acute onset of right-sided hemiparesis and speech difficulty. NCCT showed a hypodense lesion with ill-defined margins in the left insular cortex, left caudate nucleus, left internal capsule, and left frontotemporoparietal lobes, consistent with a subacute cerebral infarction in the middle cerebral artery (MCA) territory, with an ASPECTS score of 2. Due to the extensive ischemic burden and the limited availability of advanced treatment options, conservative management was chosen. The radiologist's interpretation of the ASPECTS score played a critical role in guiding the clinical team's decision-making and informing the patient's family about the prognosis. Conclusion: ASPECTS is an essential tool for predicting stroke outcomes, with lower scores correlating with larger infarct volumes and poorer prognoses. In resource-limited settings, radiologists play a vital role in interpreting ASPECTS scores to guide clinical management and provide accurate prognostic information to patients and their families.

Analysis of panel physician inquiries to U.S. TB Centers of Excellence, 2018-2022.

Applicants seeking entry into the United States are examined overseas for TB by panel physicians and international immigration clinicians guided by Centers for Disease Control and Prevention (CDC) TB Technical Instructions. To support this effort, CDC-funded TB Centers of Excellence (COEs) provide web-based expert consultation, with documentation stored in a medical consultation database (MCD). MCD analysis can reveal inquiry trends among panel physicians worldwide. TB-related queries in the COE MCD from January 1, 2018, to December 31, 2022, were analyzed using a descriptive coding scheme developed through inductive analysis, allowing multiple themes per entry. A total of 215 queries from 126 patients in 28 countries were analyzed. Major themes included evaluating diagnostic criteria, tailoring treatment, and managing comorbidities or adverse reactions. Diagnostic questions (n = 104, 48.4%) included mycobacterial culture, smear, and radiology interpretation. Treatment tailoring inquiries involved optimizing the initial regimen (n = 89, 41.4%) or modifying existing regimens (n = 26, 12.1%). Additionally, 50 consultations (23.2%) mentioned comorbidities, while 47 (21.9%) described adverse reactions. The MCD analysis identified topics where TB expertise was provided in overseas medical evaluation. These topics highlight opportunities for targeted panel physician education to improve the health of individual applicants and advance U.S. TB elimination efforts.

Gamma Dose Rate Measurements in Northern Spain: Influence of Local Meteorological Scenarios on Radiological "False Alarms" in a Real-Time Radiological Monitoring Network.

The present study characterizes gamma dose rate (GDR) concentrations in Bilbao, located in the northern Iberian Peninsula, utilizing a comprehensive 10-min interval database spanning from 2009 to 2018. This station belongs to the radiological environmental monitoring of the Basque Country network. The daily average GDR was found to be 0.07624 ± 0.00004 µSv/h, with the daily 95th percentile averaging 0.08026 ± 0.00007 µSv/h throughout the entire period. Our analysis specifically addresses the impact of precipitation on GDR, revealing a higher correlation coefficient for daily 95th percentile values compared to daily averages. Additionally, the influence of the Galerna (GL) event, a natural meteorological phenomenon in this region, on GDR was investigated, noting that it can develop both with and without precipitation. Understanding the interaction between GDR and this meteorological scenario is vital for the development of more reliable radiological monitoring strategies and for safeguarding public health. For this purpose, 40 GL events were analyzed. The present results indicate that GDR values frequently exceed alarm levels when a GL is formed. These GDR peaks should be considered natural radiological events, necessitating the classification of such GDR peaks as false alarms within the radiological monitoring network. To explain them in detail, 10-min time series of precipitation and radon outdoor concentrations were analyzed. The results demonstrate that the GL event with precipitation is a meteorological scenario that can be associated with false alarms. The present analysis provides a distinct contrast in radon behavior under the same meteorological event in case of being developed with precipitation or without precipitation. The findings from this analysis are crucial for avoiding and understanding false radiological alarms triggered in the monitoring network, thereby enhancing the accuracy of radiological data interpretation and improving public safety measures.

The Role of Artificial Intelligence in Diagnostic Radiology.

This article explores the significant impact of artificial intelligence (AI) on radiology through a comprehensive analysis of eight articles published between 2018 and 2024. With the rapid progress of modern science, the diagnostic methods in medicine are subject to change, which creates the need to consider and evaluate new diagnostic techniques such as artificial intelligence. In our study, we will evaluate the diagnostic accuracy of artificial intelligence and radiological image interpretation, as well as the pros and cons of its use and future development prospects in this field. In this article, we also consider the possibility of using GPT-4 for image analysis in radiology. Artificial intelligence is a revolutionary medical tool that can change diagnostic strategies to improve the quality of medical services.

Simulating clinical features on chest radiographs for medical image exploration and CNN explainability using a style-based generative adversarial autoencoder

Explainability of convolutional neural networks (CNNs) is integral for their adoption into radiological practice. Commonly used attribution methods localize image areas important for CNN prediction but do not characterize relevant imaging features underlying these areas, acting as a barrier to the adoption of CNNs for clinical use. We therefore propose Semantic Exploration and Explainability using a Style-based Generative Adversarial Autoencoder Network (SEE-GAAN), an explainability framework that uses latent space manipulation to generate a sequence of synthetic images that semantically visualizes how clinical and CNN features manifest within medical images. Visual analysis of changes in these sequences then facilitates the interpretation of features, thereby improving explainability. SEE-GAAN was first developed on a cohort of 26,664 chest radiographs across 15,409 patients from our institution. SEE-GAAN sequences were then generated across several clinical features and CNN predictions of NT-pro B-type natriuretic peptide (BNPP) as a proxy for acute heart failure. Radiological interpretations indicated SEE-GAAN sequences captured relevant changes in anatomical and pathological morphology associated with clinical and CNN predictions and clarified ambiguous areas highlighted by commonly used attribution methods. Our study demonstrates SEE-GAAN can facilitate our understanding of clinical features for imaging biomarker exploration and improve CNN transparency over commonly used explainability methods.

Revolution or risk?-Assessing the potential and challenges of GPT-4V in radiologic image interpretation.

ChatGPT-4 Vision (GPT-4V) is a state-of-the-art multimodal large language model (LLM) that may be queried using images. We aimed to evaluate the tool's diagnostic performance when autonomously assessing clinical imaging studies. A total of 206 imaging studies (i.e., radiography (n = 60), CT (n = 60), MRI (n = 60), and angiography (n = 26)) with unequivocal findings and established reference diagnoses from the radiologic practice of a large university hospital were accessed. Readings were performed uncontextualized, with only the image provided, and contextualized, with additional clinical and demographic information. Responses were assessed along multiple diagnostic dimensions and analyzed using appropriate statistical tests. With its pronounced propensity to favor context over image information, the tool's diagnostic accuracy improved from 8.3% (uncontextualized) to 29.1% (contextualized, first diagnosis correct) and 63.6% (contextualized, correct diagnosis among differential diagnoses) (p ≤ 0.001, Cochran's Q test). Diagnostic accuracy declined by up to 30% when 20 images were re-read after 30 and 90 days and seemed unrelated to the tool's self-reported confidence (Spearman's ρ = 0.117 (p = 0.776)). While the described imaging findings matched the suggested diagnoses in 92.7%, indicating valid diagnostic reasoning, the tool fabricated 258 imaging findings in 412 responses and misidentified imaging modalities or anatomic regions in 65 images. GPT-4V, in its current form, cannot reliably interpret radiologic images. Its tendency to disregard the image, fabricate findings, and misidentify details, especially without clinical context, may misguide healthcare providers and put patients at risk. Question Can Generative Pre-trained Transformer 4 Vision (GPT-4V) interpret radiologic images-with and without clinical context? Findings GPT-4V performed poorly, demonstrating diagnostic accuracy rates of 8% (uncontextualized), 29% (contextualized, most likely diagnosis correct), and 64% (contextualized, correct diagnosis among differential diagnoses). Clinical relevance The utility of commercial multimodal large language models, such as GPT-4V, in radiologic practice is limited. Without clinical context, diagnostic errors and fabricated findings may compromise patient safety and misguide clinical decision-making. These models must be further refined to be beneficial.

Assessment of the Diagnostic Performance of a Commercially Available Artificial Intelligence Algorithm for Risk Stratification of Thyroid Nodules on Ultrasound.

Background: Thyroid nodules are challenging to accurately characterize on ultrasound (US), though the emergence of risk stratification systems and more recently artificial intelligence (AI) algorithms has improved nodule classification. The purpose of this study was to evaluate the performance of a recent Food and Drug Administration (FDA)-cleared AI tool for detection of malignancy in thyroid nodules on US. Methods: One year of consecutive thyroid US with ≥1 nodule from Duke University Hospital and its affiliate community hospital (649 nodules from 347 patients) were retrospectively evaluated. Included nodules had ground truth diagnoses by surgical pathology, fine needle aspiration (FNA), or three-year follow-up US showing stability. An FDA-cleared AI tool (Koios DS Thyroid) analyzed each nodule to generate (i) American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) descriptors, scores, and follow-up recommendations and (ii) an AI-adapter score to further adjust risk assessments and recommendations. Four groups were then compared: (i) Koios with AI-adapter, (ii) Koios without AI-adapter, (iii) clinical radiology report, and (iv) radiology report combined with AI-adapter. Performance of the final recommendations (FNA or no FNA) was determined based on ground truth, and comparison between the four groups was made using sensitivity, specificity, and receiver-operating-curve analysis. Results: Of 649 nodules, 32 were malignant and 617 were benign. Performance of Koios with AI-adapter enabled was similar to radiologists (area under the curve [AUC] 0.70 for both, [CI 0.60-0.81] and [0.60-0.79], respectively). Koios with AI-adapter had improved specificity compared to radiologists (0.63 [CI: 0.59-0.67] versus 0.43 [CI: 0.38-0.48]) but decreased sensitivity (0.69 [CI: 0.50-0.83) versus 0.81 [CI: 0.61, 0.92]). Highest performance was seen when the radiology interpretation was combined with the AI-adapter (AUC 0.76, [CI: 0.67-0.85]). Combined with the AI-adapter, radiologist specificity improved from 0.43 ([CI: 0.38-0.48]) to 0.53 ([CI: 0.49-0.58]) (McNemar's test p < 0.001), resulting in 17% fewer FNA recommendations, with unchanged sensitivity (0.81, p = 1). Conclusion: Koios DS demonstrated standalone performance similar to radiologists, though with lower sensitivity and higher specificity. Performance was best when radiologist interpretations were combined with the AI-adapter component, with improved specificity and reduced unnecessary FNA recommendations.

Characteristics of the Reciprocal Movement of Radiographers’ Gaze Based on a Comparison of Entry-Level and Experienced Radiographers

With the development of AI, AI may substitute for physicians in radiological interpretations in the future. The purpose of this paper is to identify factors that could form the foundation for algorithms enabling AI to explore pathological findings. This study focuses on the "angles of gaze trajectory" and "average speeds of gaze movement," and conducts a comparative analysis between entry-level and experienced radiographers. Firstly, based on the results of angles of gaze movement, experienced radiographers exhibited frequent movements around the 180-degree mark compared to entry-level radiographer. Both entrylevel and experienced radiographers commonly displayed frequent movements around 0 degrees and near 350 degrees. Secondly, according to the results of average speeds of gaze movement, entry-level radiographers were found to have faster movement speeds compared to experienced radiographers.

Improvement of quality of life in women ≤ 25-years-old with chronic pelvic pain following stenting of nonthrombotic iliac vein compression.

Stenting of nonthrombotic iliac vein compression for chronic pelvic pain (CPP) is controversial, especially in young populations. The current study evaluated the effects of iliac vein stenting on CPP through the assessment of quality-of-life (QoL) and pain scores in female patients ≤25-year-old. A retrospective single site cohort study was conducted for 30 female patients who were treated with left common iliac vein (LCIV) stenting for CPP. QoL was assessed before and after stenting using the following assessments: International Pelvic Pain Society (IPPS), Pelvic Congestion Syndrome (PCS), Pelvic Pain and Urinary Urgency Frequency (PUF), Rome III IBS, ancillary symptoms assessment, and pain scores. Secondary evaluation of the accuracy of radiologic interpretations of iliac vein compression on CT and MRI was also performed. Finally, commonly associated comorbidities within this population were assessed. Patient reported outcomes in 30 iliac vein stent patients demonstrated improvements in QoL post-stenting as demonstrated through significant reductions in IPPS, PCS, and pain scores (p = .0103, .0156, and .0092, respectively). Many of the cross-sectional imaging studies that were read by the interpreting radiologist as normal or mild compression were later identified as moderate or severe by the interventional radiologist clinically evaluating the patient. These patients went on to show significant compression on venography and intravascular ultrasound with associated clinical improvement after stenting. Significant improvements in QoL were identified in young women with CPP after LCIV stenting. Similarities in the clinical presentation of common comorbidities and misinterpretation of cross-sectional imaging studies may make the identification of pelvic venous disease more challenging. Our findings endorse the need for further research on stenting in young women with venous origin CPP through comparative outcomes studies and blinded randomized controlled trials.

The “Hungry Judge” effect on prostate MRI reporting: Chronobiological trends from 35’004 radiologist interpretations

AimTo investigate the associations between the hour of the day and Prostate Imaging–Reporting and Data System (PI-RADS) scores assigned by radiologists in prostate MRI reports. Materials and methodsRetrospective single-center collection of prostate MRI reports over an 8-year period. Mean PI-RADS scores assigned between 0800 and 1800 h were examined with a regression model. ResultsA total of 35′004 prostate MRI interpretations by 26 radiologists were included. A significant association between the hour of day and mean PI-RADS score was identified (β2 = 0.005, p < 0.001), with malignant scores more frequently assigned later in the day. ConclusionThese findings suggest chronobiological factors may contribute to variability in radiological assessments. Though the magnitude of the effect is small, this may potentially add variability and impact diagnostic accuracy.

Variability in chest radiology interpretation between thoracic and non-thoracic radiologists: Implications for pulmonary fibrosis care

Variability in chest radiology interpretation between thoracic and non-thoracic radiologists: Implications for pulmonary fibrosis care

Concordance and Discordance Between Radiology Residents and Consultant Radiologist Interpretation Of CT Brain

OBJECTIVES The primary objective of this study is to assess the degree of concordance and discordance between the interpretations of computed tomography (CT) brain images by resident and consultant radiologists while emphasizing the critical significance of accurate image interpretation for informed clinical decision-making. METHODOLOGY The evaluation of radiology reports for CT Brain interpretation through a prospective analysis at the Radiology Department of Rehman Medical Institute over two years, from 1st October 2020 to 31st October 2022. A total of 198 patients who underwent cranial CT scans were interpreted by residents (R1, R2, R3, R4). Following this, the consultant radiologists reviewed the images and completed their reports. The reports of the residents and the consultant radiologists were then compared, and concordance was achieved when the residents’ reports were consistent with the final radiologist’s reports. The data collected were recorded in Microsoft Excel. The statistical analysis was performed using SPSS version 22 (IBM Corp., Armonk, NY), and the kappa coefficient was used to determine the level of agreement between residents and consultants.RESULTSAmong the 198 CT Head reports evaluated, 186 of them were in agreement with the final report of the consultant radiologist. Of the correctly diagnosed cases, R1 correctly diagnosed 46 cases, R2 correctly diagnosed 80 cases, R3 correctly diagnosed 54 cases, and R4 correctly diagnosed 6 cases. Our study achieved a percentage agreement of 93.93, with a Cohen's kappa coefficient of 0.8.CONCLUSIONThe overall concordance rate between residents and consultant radiologists was 93.93%, with a kappa coefficient 0.8. This high kappa coefficient indicates strong agreement between the two groups.

Role of teleradiology in the interpretation of ultrasound images acquired in the emergency setting

BACKGROUND: Teleradiology has become an important tool in emergency medicine, particularly in the interpretation of emergency ultrasonography. In emergency situations, where time is essential, rapid diagnosis and treatment can mean the difference between life and death. Teleradiology is an innovative alternative to augment the staffing and fill the gaps of onsite radiology personnel in emergency departments or during off-hours. AIM: To assess the effectiveness/impact of teleradiology in emergency ultrasound interpretation. MATERIALS AND METHODS: A retrospective study was performed in a cohort of 33,616 patients from 86 hospitals across the USA between January and December 2022. The study involved radiological interpretations of 37,253 ultrasound images acquired in the emergency setting by American Board Certified Radiologists empaneled by a teleradiology service provider, headquartered in Bangalore, India. RESULTS: The proposed telehealth model provided timely and quality reporting of 37,253 scans of patients with a mean turnaround time of 35.71 min (95% confidence interval 35.50–35.91). CONCLUSION: This study demonstrates that a structured telesonography program with defined protocols for image capture, transmission, and clinical communication can allow for successful immediate reporting of ultrasound data in the emergency care setting.

Application of a deep learning algorithm for the diagnosis of HCC

Background & AimsHepatocellular carcinoma (HCC) is characterized by a high mortality rate. The Liver Imaging Reporting and Data System (LI-RADS) results in considerable proportions of indeterminate observations, rendering an accurate diagnosis difficult. MethodsWe developed four deep learning models for diagnosing HCC on computed tomography (CT) via a training-validation-testing approach. Thin-slice triphasic CT liver images and relevant clinical information were collected and processed for deep learning. HCC was diagnosed and verified via a 12-month clinical composite reference standard. CT observations among at-risk patients were annotated using LI-RADS. Diagnostic performance was assessed by internal validation and independent external testing. We conducted sensitivity analyses of different subgroups, deep learning explainability evaluation, and misclassification analysis. ResultsFrom 2,832 patients and 4,305 CT observations, the best-performing model was Spatio-Temporal 3D Convolution Network (ST3DCN), achieving area under curves (AUCs) of 0.919 (95%CI 0.903-0.935) and 0.901 (95%CI 0.879-0.924) at the observation (n=1077) and patient (n=685) levels respectively during internal validation, compared to 0.839 (95%CI 0.814-0.864) and 0.822 (95%CI 0.790-0.853) respectively for standard-of-care radiological interpretation. ST3DCN’s negative predictive values were 0.966 (95%CI 0.954-0.979) and 0.951 (95%CI 0.931-0.971) respectively. ST3DCN’s observation-level AUCs among at-risk patients, 2-5 cm observations and singular porto-venous phase analysis were 0.899 (95%CI 0.874-0.924), 0.872 (95%CI 0.838-0.909) and 0.912 (95%CI 0.895-0.929) respectively. In external testing (551/717 patients/observations), ST3DCN’s AUC was 0.901 (95%CI 0.877-0.924), non-inferior to radiological interpretation (AUC 0.900, 95%CI 0.877-923). ConclusionsST3DCN achieved strong, robust performance for accurate HCC diagnosis on CT. Deep learning can expedite and improve the diagnostic process of HCC. Impact and implicationsThe clinical applicability of deep learning in HCC diagnosis is potentially huge, especially considering the expected increase in the incidence and mortality of HCC in Eastern Asia and worldwide. Early diagnosis through deep learning can lead to earlier definitive management, particularly for at-risk patients. The model can be broadly deployed for patients undergoing a triphasic contrast CT scan of the liver to reduce the current high mortality rate of HCC.