Expert Radiologists Research Articles

COVID-19 Infection Percentage Estimation from Computed Tomography Scans: Results and Insights from the International Per-COVID-19 Challenge.

COVID-19 analysis from medical imaging is an important task that has been intensively studied in the last years due to the spread of the COVID-19 pandemic. In fact, medical imaging has often been used as a complementary or main tool to recognize the infected persons. On the other hand, medical imaging has the ability to provide more details about COVID-19 infection, including its severity and spread, which makes it possible to evaluate the infection and follow-up the patient's state. CT scans are the most informative tool for COVID-19 infection, where the evaluation of COVID-19 infection is usually performed through infection segmentation. However, segmentation is a tedious task that requires much effort and time from expert radiologists. To deal with this limitation, an efficient framework for estimating COVID-19 infection as a regression task is proposed. The goal of the Per-COVID-19 challenge is to test the efficiency of modern deep learning methods on COVID-19 infection percentage estimation (CIPE) from CT scans. Participants had to develop an efficient deep learning approach that can learn from noisy data. In addition, participants had to cope with many challenges, including those related to COVID-19 infection complexity and crossdataset scenarios. This paper provides an overview of the COVID-19 infection percentage estimation challenge (Per-COVID-19) held at MIA-COVID-2022. Details of the competition data, challenges, and evaluation metrics are presented. The best performing approaches and their results are described and discussed.

Evaluation of retrieval accuracy and visual similarity in content-based image retrieval of chest CT for obstructive lung disease

The aim of our study was to assess the performance of content-based image retrieval (CBIR) for similar chest computed tomography (CT) in obstructive lung disease. This retrospective study included patients with obstructive lung disease who underwent volumetric chest CT scans. The CBIR database included 600 chest CT scans from 541 patients. To assess the system performance, follow-up chest CT scans of 50 patients were evaluated as query cases, which showed the stability of the CT findings between baseline and follow-up chest CT, as confirmed by thoracic radiologists. The CBIR system retrieved the top five similar CT scans for each query case from the database by quantifying and comparing emphysema extent and size, airway wall thickness, and peripheral pulmonary vasculatures in descending order from the database. The rates of retrieval of the same pairs of query CT scans in the top 1–5 retrievals were assessed. Two expert chest radiologists evaluated the visual similarities between the query and retrieved CT scans using a five-point scale grading system. The rates of retrieving the same pairs of query CTs were 60.0% (30/50) and 68.0% (34/50) for top-three and top-five retrievals. Radiologists rated 64.8% (95% confidence interval 58.8–70.4) of the retrieved CT scans with a visual similarity score of four or five and at least one case scored five points in 74% (74/100) of all query cases. The proposed CBIR system for obstructive lung disease integrating quantitative CT measures demonstrated potential for retrieving chest CT scans with similar imaging phenotypes. Further refinement and validation in this field would be valuable.

Deep learning performance on MRI prostate gland segmentation: evaluation of two commercially available algorithms compared with an expert radiologist.

Accurate whole-gland prostate segmentation is crucial for successful ultrasound-MRI fusion biopsy, focal cancer treatment, and radiation therapy techniques. Commercially available artificial intelligence (AI) models, using deep learning algorithms (DLAs) for prostate gland segmentation, are rapidly increasing in numbers. Typically, their performance in a true clinical context is scarcely examined or published. We used a heterogenous clinical MRI dataset in this study aiming to contribute to validation of AI-models. We included 123 patients in this retrospective multicenter (7 hospitals), multiscanner (8 scanners, 2 vendors, 1.5T and 3T) study comparing prostate contour assessment by 2 commercially available Food and Drug Association (FDA)-cleared and CE-marked algorithms (DLA1 and DLA2) using an expert radiologist's manual contours as a reference standard (RSexp) in this clinical heterogeneous MRI dataset. No in-house training of the DLAs was performed before testing. Several methods for comparing segmentation overlap were used, the Dice similarity coefficient (DSC) being the most important. The DSC mean and standard deviation for DLA1 versus the radiologist reference standard (RSexp) was and for DLA2 versus RSexp it was . A paired -test to compare the DSC for DLA1 and DLA2 showed no statistically significant difference (). Two commercially available DL algorithms (FDA-cleared and CE-marked) can perform accurate whole-gland prostate segmentation on a par with expert radiologist manual planimetry on a real-world clinical dataset. Implementing AI models in the clinical routine may free up time that can be better invested in complex work tasks, adding more patient value.

Automatic generation of conclusions from neuroradiology MRI reports through natural language processing.

The conclusion section of a radiology report is crucial for summarizing the primary radiological findings in natural language and essential for communicating results to clinicians. However, creating these summaries is time-consuming, repetitive, and prone to variability and errors among different radiologists. To address these issues, we evaluated a fine-tuned Text-To-Text Transfer Transformer (T5) model for abstractive summarization to automatically generate conclusions for neuroradiology MRI reports in a low-resource language. We retrospectively applied our method to a dataset of 232,425 neuroradiology MRI reports in Spanish. We compared various pre-trained T5 models, including multilingual T5 and those newly adapted for Spanish. For precise evaluation, we employed BLEU, METEOR, ROUGE-L, CIDEr, and cosine similarity metrics alongside expert radiologist assessments. The findings are promising, with the models specifically fine-tuned for neuroradiology MRI achieving scores of 0.46, 0.28, 0.52, 2.45, and 0.87 in the BLEU-1, METEOR, ROUGE-L, CIDEr, and cosine similarity metrics, respectively. In the radiological experts' evaluation, they found that in 75% of the cases evaluated, the conclusions generated by the system were as good as or even better than the manually generated conclusions. The methods demonstrate the potential and effectiveness of customizing state-of-the-art pre-trained models for neuroradiology, yielding automatic MRI report conclusions that nearly match expert quality. Furthermore, these results underscore the importance of designing and pre-training a dedicated language model for radiology report summarization.

Utilization of texture features of volumetric ADC maps in differentiating between serous cystadenoma and intraductal papillary neoplasms.

The rising incidence of incidental detection of pancreatic cystic neoplasms has compelled radiologists to determine new diagnostic methods for the differentiation of various kinds of lesions. We aim to demonstrate the utility of texture features extracted from ADC maps in differentiating intraductal papillary mucinous neoplasms (IPMN) from serous cystadenomas (SCA). This retrospective study was performed on 136 patients (IPMN = 87, SCA = 49) split into testing and training datasets. A total of 851 radiomics features were extracted from volumetric contours drawn by an expert radiologist on ADC maps of the lesions. LASSO regression analysis was used to determine the most predictive set of features and a radiomics score was developed based on their respective coefficients. A hyper-optimized support vector machine was then utilized to classify the lesions based on their radiomics score. A total of four Wavelet features (LHL/GLCM/LCM2, HLL/GLCM/LCM2, /LLL/First Order/90percent, /LLL/GLCM/MCC) were selected from all of the features to be included in our classifier. The classifier was optimized by altering hyperparameters and the trained model was applied to the validation dataset. The model achieved a sensitivity of 92.8, specificity of 90%, and an AUC of 0.97 in the training data set, and a sensitivity of 83.3%, specificity of 66.7%, and AUC of 0.90 in the testing dataset. A support vector machine model trained and validated on volumetric texture features extracted from ADC maps showed the possible beneficence of these features in differentiating IPMNs from SCAs. These results are in line with previous regarding the role of ADC maps in classifying cystic lesions and offers new evidence regarding the role of texture features in differentiation of potentially neoplastic and benign lesions.

Strain Elastography Fat-to-Lesion Index Is Associated with Mammography BI-RADS Grading, Biopsy, and Molecular Phenotype in Breast Cancer.

Breast cancer (BC) affects millions of women worldwide, causing over 500,000 deaths annually. It is the leading cause of cancer mortality in women, with 70% of deaths occurring in developing countries. Elastography, which evaluates tissue stiffness, is a promising real-time minimally invasive technique for BC diagnosis. This study assessed strain elastography (SE) and the fat-to-lesion (F/L) index for BC diagnosis. This prospective study included 216 women who underwent SE, ultrasound, mammography, and breast biopsy (108 malignant, 108 benign). Three expert radiologists performed imaging and biopsies. Mean F/L index was 3.70 ± 2.57 for benign biopsies and 18.10 ± 17.01 for malignant. We developed two predictive models: a logistic regression model with AUC 0.893, 79.63% sensitivity, 87.62% specificity, 86.9% positive predictive value (+PV), and 80.7% negative predictive value (-PV); and a neural network with AUC 0.902, 80.56% sensitivity, 88.57% specificity, 87.9% +PV, and 81.6% -PV. The optimal Youden F/L index cutoff was >5.76, with 84.26% sensitivity and specificity. The F/L index positively correlated with BI-RADS (Spearman's r = 0.073, p < 0.001) and differed among molecular subtypes (Kruskal-Wallis, p = 0.002). SE complements mammography for BC diagnosis. With adequate predictive capacity, SE is fast, minimally invasive, and useful when mammography is contraindicated.

4. ULTRASOUND STUDY OF TROCAR HERNIA IN POST-OPERATED PATIENTS OF BARIATRIC SURGERY: MEDIUM AND LONG-TERM FOLLOW-UP

Abstract Introduction Patients undergoing bariatric surgery have comorbidities and risk factors that increase the likelihood of incisional hernias. Bladeless trocars are safe and not very harmful, so we opted not to close them due to the difficulty they present in these patients. Methods Observational, descriptive, longitudinal, and retrospective study in patients who underwent bariatric surgery by laparoscopy with 12mm trocars without a blade or fascial closure between January 2015 and July 2016. Data on risk factors, pain scale, and ultrasound of ports by an expert radiologist in 2018 and 2022. Results A total of 45 patients, with a mean age of 49.9 years and an initial BMI of 45.5 kg/m², were included in the study. The majority (88.9%) underwent gastric bypass, while the rest underwent vertical gastrectomy. In 2018, 45 abdominal wall ultrasounds were conducted, revealing 7 trocar hernias out of 225 explored ports (3.1%). All hernias were located in the epigastric port and were asymptomatic. In 2022, 32 ultrasound scans were performed with findings of 7 hernias (4.4%), 3 new and 4 known, with 2 symptomatic patients at the site of the hernia with mild symptoms not consulted. Conclusion The incidence of trocar hernia in the medium and long term following bariatric surgery with 12mm ports, without blade or fascial closure, is notably low, with all cases located in the midline (p&amp;lt;0.00001). These findings lead us to question the necessity of closing transmuscular accesses and assess the closure or lateralization of the midline trocar.

Lung imaging patterns in connective tissue disease-associated interstitial lung disease impact prognosis and immunosuppression response.

Interstitial lung disease (ILD) in CTDs has highly variable morphology. We aimed to identify imaging features and their impact on ILD progression, mortality, and immunosuppression response. Patients with CTD-ILD had high-resolution chest CT (HRCT) reviewed by expert radiologists blinded to clinical data for overall imaging pattern [usual interstitial pneumonia (UIP); non-specific interstitial pneumonia (NSIP); organizing pneumonia (OP); fibrotic hypersensitivity pneumonitis (fHP); and other]. Transplant-free survival and change in percent-predicted forced vital capacity (FVC) were compared using Cox and linear mixed-effects models adjusted for age, sex, smoking, and baseline FVC. FVC decline after immunosuppression was compared with pre-treatment. Among 645 CTD-ILD patients, the most frequent CTDs were SSc (n = 215), RA (n = 127), and inflammatory myopathies (n = 100). NSIP was the most common pattern (54%), followed by UIP (20%), fHP (9%), and OP (5%). Compared with the case for patients with UIP, FVC decline was slower in patients with NSIP (by 1.1%/year, 95% CI 0.2, 1.9) or OP (by 3.5%/year, 95% CI 2.0, 4.9), and mortality was lower in patients with NSIP [hazard ratio (HR) 0.65, 95% CI 0.45, 0.93] or OP (HR 0.18, 95% CI 0.05, 0.57), but higher in fHP (HR 1.58, 95% CI 1.01, 2.40). The extent of fibrosis also predicted FVC decline and mortality. After immunosuppression, FVC decline was slower compared with pre-treatment in NSIP (by 2.1%/year, 95% CI 1.4, 2.8), with no change for UIP or fHP. Multiple radiologic patterns are possible in CTD-ILD, including a fHP pattern. NSIP and OP were associated with better outcomes and response to immunosuppression, while fHP had worse survival compared with UIP.

Technical and Clinical Outcome of Low-Milliampere CT Fluoroscopy-Guided Percutaneous Drainage Placement in Abdominal Fluid Collections after Liver Transplantation: A 16-Year Retrospective Analysis of 50 Consecutive Patients.

Evaluation of the effectiveness of CT-guided drainage (CTD) placement in managing symptomatic postoperative fluid collections in liver transplant patients. The assessment included technical success, clinical outcomes, and the occurrence of complications during the peri-interventional period. Analysis spanned the years 2005 to 2020 and involved 91 drain placement sessions in 50 patients using percutaneous transabdominal or transhepatic access. Criteria for technical success (TS) included (a) achieving adequate drainage of the fluid collection and (b) the absence of peri-interventional complications necessitating minor or prolonged hospitalization. Clinical success (CS) was characterized by (a) a reduction or normalization of inflammatory blood parameters within 30 days after CTD placement and (b) the absence of a need for surgical revision within 60 days after the intervention. Inflammatory markers in terms of C-reactive protein (CRP), leukocyte count and interleukin-6, were evaluated. The dose length product (DLP) for various intervention steps was calculated. The TS rate was 93.4%. CS rates were 64.3% for CRP, 77.8% for leukocytes, and 54.5% for interleukin-6. Median time until successful decrease was 5.0 days for CRP and 3.0 days for leukocytes and interleukin-6. Surgical revision was not necessary in 94.0% of the cases. During the second half of the observation period, there was a trend (p = 0.328) towards a lower DLP for the entire intervention procedure (median: years 2013 to 2020: 623.0 mGy·cm vs. years 2005 to 2012: 811.5 mGy·cm). DLP for the CT fluoroscopy component was significantly (p = 0.001) lower in the later period (median: years 2013 to 2020: 31.0 mGy·cm vs. years 2005 to 2012: 80.5 mGy·cm). The TS rate of CT-guided drainage (CTD) placement was notably high. The CS rate ranged from fair to good. The reduction in radiation exposure over time can be attributed to advancements in CT technology and the growing expertise of interventional radiologists.

A validation of register-derived diagnoses of interstitial lung disease in patients with inflammatory arthritis: data from the NOR-DMARD study

Objective There is a lack of knowledge concerning the validity of the interstitial lung disease (ILD) diagnoses used in epidemiological studies on rheumatic diseases. This paper seeks to verify register-derived ILD diagnoses using chest computed tomography (CT) and medical records as a gold standard. Method The Norwegian Anti-Rheumatic Drug Register (NOR-DMARD) is a multicentre prospective observational study of patients with inflammatory arthritis who start treatment with disease-modifying anti-rheumatic drugs. NOR-DMARD is linked to the Norwegian Patient Registry (NPR) and Cause of Death Registry. We searched registers for ILD coded by ICD-10 J84 or J99 among patients with rheumatoid arthritis, psoriatic arthritis, or spondyloarthritis. We extracted chest CT reports and medical records from participating hospitals. Two expert thoracic radiologists scored examinations to confirm the ILD diagnosis. We also searched medical records to find justifications for the diagnosis following multidisciplinary evaluations. We calculated the positive predictive values (PPVs) for ILD across subsets. Results We identified 71 cases with an ILD diagnosis. CT examinations were available in 65/71 patients (91.5%), of whom ILD was confirmed on CT in 29/65 (44.6%). In a further 10 patients, medical records confirmed the diagnosis, giving a total of 39/71 verified cases. The PPV of a register-derived ILD diagnosis was thus 54.9%. In a subset of patients who had received an ILD code at two or more time-points and had a CT scan taken within a relevant period, the PPV was 72.2%. Conclusion The validity of register-based diagnoses of ILD must be carefully considered in epidemiological studies.

Aided Diagnosis Model Based on Deep Learning for Glioblastoma, Solitary Brain Metastases, and Primary Central Nervous System Lymphoma with Multi-Modal MRI.

(1) Background: Diagnosis of glioblastoma (GBM), solitary brain metastases (SBM), and primary central nervous system lymphoma (PCNSL) plays a decisive role in the development of personalized treatment plans. Constructing a deep learning classification network to diagnose GBM, SBM, and PCNSL with multi-modal MRI is important and necessary. (2) Subjects: GBM, SBM, and PCNSL were confirmed by histopathology with the multi-modal MRI examination (study from 1225 subjects, average age 53 years, 671 males), 3.0 T T2 fluid-attenuated inversion recovery (T2-Flair), and Contrast-enhanced T1-weighted imaging (CE-T1WI). (3) Methods: This paper introduces MFFC-Net, a classification model based on the fusion of multi-modal MRIs, for the classification of GBM, SBM, and PCNSL. The network architecture consists of parallel encoders using DenseBlocks to extract features from different modalities of MRI images. Subsequently, an L1-norm feature fusion module is applied to enhance the interrelationships among tumor tissues. Then, a spatial-channel self-attention weighting operation is performed after the feature fusion. Finally, the classification results are obtained using the full convolutional layer (FC) and Soft-max. (4) Results: The ACC of MFFC-Net based on feature fusion was 0.920, better than the radiomics model (ACC of 0.829). There was no significant difference in the ACC compared to the expert radiologist (0.920 vs. 0.924, p = 0.774). (5) Conclusions: Our MFFC-Net model could distinguish GBM, SBM, and PCNSL preoperatively based on multi-modal MRI, with a higher performance than the radiomics model and was comparable to radiologists.

Machine learning-based identification of contrast-enhancement phase of computed tomography scans.

Contrast-enhanced computed tomography scans (CECT) are routinely used in the evaluation of different clinical scenarios, including the detection and characterization of hepatocellular carcinoma (HCC). Quantitative medical image analysis has been an exponentially growing scientific field. A number of studies reported on the effects of variations in the contrast enhancement phase on the reproducibility of quantitative imaging features extracted from CT scans. The identification and labeling of phase enhancement is a time-consuming task, with a current need for an accurate automated labeling algorithm to identify the enhancement phase of CT scans. In this study, we investigated the ability of machine learning algorithms to label the phases in a dataset of 59 HCC patients scanned with a dynamic contrast-enhanced CT protocol. The ground truth labels were provided by expert radiologists. Regions of interest were defined within the aorta, the portal vein, and the liver. Mean density values were extracted from those regions of interest and used for machine learning modeling. Models were evaluated using accuracy, the area under the curve (AUC), and Matthew's correlation coefficient (MCC). We tested the algorithms on an external dataset (76 patients). Our results indicate that several supervised learning algorithms (logistic regression, random forest, etc.) performed similarly, and our developed algorithms can accurately classify the phase of contrast enhancement.

Artificial intelligence (AI)-assisted chest computer tomography (CT) insights: a study on intensive care unit (ICU) admittance trends in 78 coronavirus disease 2019 (COVID-19) patients.

The global coronavirus disease 2019 (COVID-19) pandemic has posed substantial challenges for healthcare systems, notably the increased demand for chest computed tomography (CT) scans, which lack automated analysis. Our study addresses this by utilizing artificial intelligence-supported automated computer analysis to investigate lung involvement distribution and extent in COVID-19 patients. Additionally, we explore the association between lung involvement and intensive care unit (ICU) admission, while also comparing computer analysis performance with expert radiologists' assessments. A total of 81 patients from an open-source COVID database with confirmed COVID-19 infection were included in the study. Three patients were excluded. Lung involvement was assessed in 78 patients using CT scans, and the extent of infiltration and collapse was quantified across various lung lobes and regions. The associations between lung involvement and ICU admission were analysed. Additionally, the computer analysis of COVID-19 involvement was compared against a human rating provided by radiological experts. The results showed a higher degree of infiltration and collapse in the lower lobes compared to the upper lobes (P<0.05). No significant difference was detected in the COVID-19-related involvement of the left and right lower lobes. The right middle lobe demonstrated lower involvement compared to the right lower lobes (P<0.05). When examining the regions, significantly more COVID-19 involvement was found when comparing the posterior vs. the anterior halves and the lower vs. the upper half of the lungs. Patients, who required ICU admission during their treatment exhibited significantly higher COVID-19 involvement in their lung parenchyma according to computer analysis, compared to patients who remained in general wards. Patients with more than 40% COVID-19 involvement were almost exclusively treated in intensive care. A high correlation was observed between computer detection of COVID-19 affections and the rating by radiological experts. The findings suggest that the extent of lung involvement, particularly in the lower lobes, dorsal lungs, and lower half of the lungs, may be associated with the need for ICU admission in patients with COVID-19. Computer analysis showed a high correlation with expert rating, highlighting its potential utility in clinical settings for assessing lung involvement. This information may help guide clinical decision-making and resource allocation during ongoing or future pandemics. Further studies with larger sample sizes are warranted to validate these findings.

Histogram-Based Densitometry Index to Assess the Severity of Interstitial Lung Disease in Systemic Sclerosis in Standard and Low-Dose Computed Tomography.

Mean lung attenuation, skewness, and kurtosis are histogram-based densitometry variables that quantify systemic sclerosis-associated interstitial lung disease (SSc-ILD) and were recently merged into a computerized integrated index (CII). Our work tested the CII in low-dose 9-slice (reduced) and standard high-resolution computed tomography (CT) scans to evaluate extensive SSc-ILD and predict mortality. CT scans from patients with SSc-ILD were assessed using the software Horos to compute standard and reduced CIIs. Extensive ILD was determined following the Goh staging system. The association between CIIs and extensive ILD was analyzed with a generalized estimating equation regression model, the predictive ability of CIIs by the area under the receiver-operation characteristic curve (AUC), and the association between CIIs and death by Kaplan-Meier analysis. Among 243 patients with standard and reduced CT scans available, 157 CT scans from 119 patients with SSc-ILD constituted the derivation cohort. The validation cohort included 116 standard and 175 reduced CT scans. Both CIIs from standard (odds ratio [OR] 0.53, 95% CI 0.37-0.75; AUC 0.77, 95% CI 0.68-0.87) and reduced CT scans (OR 0.54, 95% CI 0.35-0.82; AUC 0.78, 95% CI 0.70-0.87) were significantly associated with extensive ILD. A threshold of CII ≤ -0.96 for standard CT scans and CII ≤ -1.85 for reduced CT scans detected extensive ILD with high sensitivity in both derivation and validation cohorts. Extensive ILD according to Goh staging (OR 2.94, 95% CI 1.10-7.82) and standard CII ≤ -0.96 (OR 1.78, 95% CI 1.24-2.56) significantly predicted mortality; a marginal P value was observed for reduced CII ≤ -1.85 (OR 1.27, 95% CI 0.93-1.75). Thresholds for both standard and reduced CII to identify extensive ILD were developed and validated, with an additional association with mortality. CIIs might help in clinical practice when radiology expertise is missing.

Utility of artificial intelligence for detection of pneumothorax on chest radiopgraphs done after transthoracic percutaneous transthoracic biopsy guided by computed tomography

ObjetiveTo assess the ability of an artificial intelligence software to detect pneumothorax in chest radiographs done after percutaneous transthoracic biopsy. Material and methodsWe included retrospectively in our study adult patients who underwent CT-guided percutaneous transthoracic biopsies from lung, pleural or mediastinal lesions from June 2019 to June 2020, and who had a follow-up chest radiograph after the procedure. These chest radiographs were read to search the presence of pneumothorax independently by an expert thoracic radiologist and a radiodiagnosis resident, whose unified lecture was defined as the gold standard, and the result of each radiograph after interpretation by the artificial intelligence software was documented for posterior comparison with the gold standard. ResultsA total of 284 chest radiographs were included in the study and the incidence of pneumothorax was 14.4%. There were no discrepancies between the two readers’ interpretation of any of the postbiopsy chest radiographs. The artificial intelligence software was able to detect 41/41 of the present pneumothorax, implying a sensitivity of 100% and a negative predictive value of 100%, with a specificity of 79.4% and a positive predictive value of 45%. The accuracy was 82.4%, indicating that there is a high probability that an individual will be adequately classified by the software. It has also been documented that the presence of Port-a-cath is the cause of 8 of the 50 of false positives by the software. ConclusionsThe software has detected 100% of cases of pneumothorax in the postbiopsy chest radiographs. A potential use of this software could be as a prioritisation tool, allowing radiologists not to read immediately (or even not to read) chest radiographs classified as non-pathological by the software, with the confidence that there are no pathological cases.

Abstract A088: Utilizing Machine Learning Techniques to Investigate Mammograms for Breast Cancer Detection

Abstract Purpose: The prevalence of artificial intelligence (AI) in breast cancer screening and diagnosis has substantially increased over the past 5 years. AI can be used to assist in the interpretation of mammograms and other breast imaging modalities, as well as predict patient outcomes. This study aims to quantify AI's diagnostic accuracy and its ability to detect breast cancer from a large database of mammograms. The implementation of AI in breast cancer screening holds immense potential to not only enhance the accuracy of mammogram interpretation but also improve the overall efficiency of radiologists and healthcare providers. Furthermore, we aim to contribute to the advancement of the rapidly growing field of machine learning in medicine, particularly in breast cancer care. Our hope is that by applying this technology, we can improve patient outcomes and increase accessibility to breast cancer screening and early detection. Methods: A deep neural network (DNN) model was built on Tensorflow utilizing Radiological Society of North America Screening Mammography Breast Cancer Detection dataset, consisting of normal mammograms and abnormal mammograms with tumors. The model's efficacy was assessed using Area under the curve (AUC), precision, and recall, with images randomly split into training (80%), validation (10%), and test sets (10%). Four consecutive training sessions each lasting 2 hours and 13 minutes using 8,877 images consisting of 4,621 breast cancer and 4,456 normal breast mammogram images was done. Results: The model attained an AUC of 0.926, Specificity of 95%, Sensitivity of 73% and Accuracy of 84%. Conclusion: This DNN model was created for diagnosing breast cancer from mammograms, with higher AUC than the current standard. This study highlights the potential of AI to revolutionize breast cancer screening, prevention, and diagnosis, which will ultimately improve patient outcomes and increase accessibility for patients. The use of AI in detecting breast cancer on mammograms can not only provide support to radiologists but also improve their efficacy and accuracy, reducing the burden of high volume images. Additionally, the use of AI in detection and screening can address the barriers that come with physician shortages, especially in underserved areas. Extending AI’s reach to these areas of limited access can improve early detection and screening which tends to be lacking in under-resourced areas, thus addressing the gap in health equity worldwide. While this study highlights encouraging data at the intersection of AI and medicine, we acknowledge that AI’s role in medicine is not to replace but rather enhance the expertise of radiologists. Lastly, we hope that our developed tool can alleviate the workload of radiologists and increase efficiency in high-pressure settings such as emergency departments and rural areas, ultimately leading to improved patient care. Citation Format: Parsa Riazi Esfahani, Maya M Maalouf, Akshay J Reddy, Prashant Chawla. Utilizing Machine Learning Techniques to Investigate Mammograms for Breast Cancer Detection [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Breast Cancer Research; 2023 Oct 19-22; San Diego, California. Philadelphia (PA): AACR; Cancer Res 2024;84(3 Suppl_1):Abstract nr A088.

P273 Defining the prognosis of children with Crohn Disease with the support of magnetic resonance enterography: a multi-center multi-reader study

Abstract Background Ileocolonoscopic assessment is the most accurate method for detecting intestinal inflammation in Crohn disease (CD), but it allows the assessment only of a limited tract of the small intestine and does not provide data on inflammation beyond the mucosa. The auxiliary use of magnetic resonance enterography (MRE) can overcome these limitations and allow to predict the course of CD. Paediatric Inflammatory Crohn's MRE Index (PICMI) is a multi-point index of intestinal inflammation (mucosal and transmural) in children with CD created on the basis of MRI. The aim of the present study is to assess whether this completely non-invasive index at diagnosis could predict the course of CD in children and to test the inter-reader agreement as surrogate of index reproducibility. Methods All children diagnosed with CD in two tertiary referral centres with a 1-year minimum follow-up were enrolled in the study. MRE at diagnosis was blindly evaluated by 3 expert radiologists and PICMI was calculated for all patients. Children’s disease was stratified according to PICMI at diagnosis into remission (&amp;lt;10), mild (11-55), moderate (56-120) and severe (&amp;gt;120). Inter-observer agreement among radiologist was calculated. Data at follow-up were collected at 6-8 weeks, 1 year, 3 years and 5 years after CD diagnosis. Association between PICMI at diagnosis and CD prognosis was evaluated. Results A total of 71 children (52 males) with proven CD were enrolled in the study and PICMI was calculated for 68. PICMI at diagnosis was stratified into remission 6 (8.8%), mild 29 (42.6%), moderate 24 (35.3%), severe 9 (13.2%). Inter-observer agreement was calculated with an intraclass correlation coefficient of 0.65 which indicates moderate reliability between the 3 raters. PICMI score at diagnosis significantly correlated with PCDAI (Paediatric Crohn's Disease Activity Index) at diagnosis (p: 0.036). Steroid-free remission at 1, 3 and 5 years was comparable between PICMI groups: remission + mild vs moderate + severe (p: 0.606). CRP and calprotectin negative steroid free remission at 1,3,5 was also comparable (p: 0.578). PICMI at diagnosis was associated with the probability of biologic introduction at 1 year: incidence rate ratio IRR: 2.17 (1.09-4.42); p=0.019, 3-year IRR: 2.12 (1.15-3.96); p=0.011, and 5 years: 2.21 (1.20-4.08); p=0.007 (figure 1). Conclusion PICMI score is a reliable and reproducible index to determine activity in children with Crohn’s disease. Children with more active disease at diagnosis according to PICMI have shown an increased need to progression to biologic treatment in order to gain a comparable rate of steroid free remission and inflammatory marker negativity if compared with children with lower PICMI scores.

P013 Magnetic Resonance Imaging-Based Average T2 Signal as A Non-Invasive Measure of Fibrosis In Enteric Crohn’s Disease

Abstract Background In Crohn’s Disease (CD), 10% of patients present with fibrostenosis with a further 10% progressing from inflammatory to a fibrostenotic disease behaviour over 7 years. 40% of fibrostenotic CD patients need surgery at 5 years compared to ~10% with inflammatory disease. Present non-invasive fibrosis biomarkers lack the required accuracy needed to be used routinely for clinical decision making. T1, T2, diffusion weighted imaging, motility and bowel volumes have shown promise in the non-invasive measure of fibrosis. Methods The primary objective of this study is to undertake biological validation of MRI measures as independent imaging biomarkers of histological fibrosis with the primary outcome being the correlation between MRI measures and histological fibrosis measures. GI-Seg (Motilent, London, UK) is an image analysis tool used to segment the gastrointestinal tract on T2W MRI scans. Patients with CD histological diagnosis and in need of surgical resection for stenosis attended a one-hour 3T MRI scanning session within 12 weeks prior to their intestinal resection surgery. Histological samples were scored by a specialized gastrointestinal pathologist for inflammation, fibrosis, and muscular hypertrophy. An expert radiologist with the help of the histopathologist located the strictures on DWI, quantitative T2&amp;T2 scans, and T2 weighted clinical MRI scans. GI-Seg was used to generate a 3D model of the stricture on T2W scans measuring bowel wall volume and average signal intensity of each stricture. Pearson correlation coefficient was used to quantify the correlation between the variables. Results A total of 10 CD patients, encompassing 20 strictures, were included in the study. The analysis revealed that the average T2 signal has a significant correlation with global fibrosis scores (r(20) = .563, p = .010) (Figure) and total inflammation scores (r(20) = .451, p = .046). Additionally, stricture volume showed a fair correlation with chronic inflammation component scores (r(20) = .453, p = .045). Notably, quantitative T1 values correlated with total inflammation scores (r(17) = .493, p = .045). Conclusion The study reveals fair correlations between the T2 signal average and both global fibrosis and global inflammation scores, as well as between stricture volume and chronic inflammation in CD strictures. While these findings indicate that the GI-seg tool's measurements of bowel wall volume and signal intensity have potential utility as imaging biomarkers for fibrosis and inflammation, the results also underscore the necessity for additional markers that are specifically correlated with fibrosis. Such markers are essential to distinctly identify fibrotic changes, independent of inflammatory processes.

P362 Mesenteric fat proliferation and fat edema in magnetic resonance enterography as predictor of long-term disease complications in Crohn’s disease

Abstract Background Fibrofatty proliferation of the mesenteric fat is a common finding in magnetic resonance enterography (MRE) in patients with Crohn’s disease (CD). Mesenteric fat is thought to play an active role in inflammation, fibrosis and stricture formation in CD. However, it is unclear whether these findings have a prognostic value in prediction of bowel damage and complications. We aimed to examine the association between mesenteric fat proliferation and inflammation on MRE and risk for hospital admissions and bowel surgery. Methods Patients with established CD diagnosis with at least 1 available MRE imaging data were followed for hospital admissions for CD exacerbation or abdominal surgery. MRE were reviewed by an expert radiologist for various characteristics (intestinal wall thickening and enhancement, stricture, or abscess formation, fibrofatty proliferation and fat edema reflecting fat inflammation). The association between mesenteric fat and outcomes was assessed using the Kaplan Meier method and log rank test. Crude and adjusted Hazard ratios (HR) with 95% confidence interval (CI) adjusting for baseline MRE characteristics (wall thickness and stricture and pre-stenotic dilatation) were estimated from the Cox proportional hazards regression model (R version 4.2.1). Results 159 CD patient charts and MRE imaging data were analyzed. Patients’ characteristics and medication exposure at described in table 1. On initial MRE the frequency of mesenteric fat proliferation was noted in 43 (27.0%) and fat edema in 79 (49.7%). Other MRE features noted were: mild, moderate and severe wall thickening in 72 (45.5%), 47 (29.9%) and 11 (7.0%) patients, respectively. Contrast intestinal wall enhancement was present in 122 (76.7%) patients. Intestinal stricture with pre-stenotic dilatation was noted in 17 (10.7%) patients and abscess and fistula formation was seen in 10 (6.3%) and 11 (6.9%) patients, respectively. During a mean (SD) follow up of 95.7 (39.5) months, 100 (62.9%) and 40 (25.2%) patients underwent hospitalization or surgery, respectively. On survival analysis, fat proliferation was associated with hospitalizations (Fig 1A, p=0.0005) and surgery (fig 1B, p&amp;lt;0.0001), while fat edema was associated with abdominal surgery (Fig 1D p=0.0003) but not with hospitalization (Fig 1C p=0.28). On multivariable analysis of the associations between fat proliferation and surgery and hospitalization remained highly significant (adjusted HR (95%CI) =3.15 (1.57-6.33) and adjusted HR (95%CI) =1.77 (1.06-2.97), respectively. Conclusion in this analysis of baseline MRE features, mesenteric fat proliferation was independently associated with increased risk for abdominal surgery and hospital admission for disease flares.

Inter-reader variability of imaging biomarkers for therapeutic response in pancreatic ductal adenocarcinoma.

700 Background: Indeterminate borders of pancreatic ductal adenocarcinoma (PDAC) can impair reliability of lesion annotation in imaging assessment of therapeutic response (TR) after neoadjuvant chemotherapy (NAT). Previous work showed that longitudinal image registration algorithms can track treatment-related changes and produce quantitative imaging biomarkers (QIB) for TR that predict overall (OS) and recurrence-free survival (RFS). We compare inter-reader agreements of these biomarkers against conventional RECIST and tumor volume QIB. Methods: N=30 patients enrolled in a Phase II clinical trial comparing outcomes of a NAT regimen were retrospectively analyzed. In baseline and restaging CT, whole pancreas and tumor segmentations and tumor diameters were annotated by 2 expert attending radiologists and 2 novice radiology fellows. For each reader, percent change in tumor diameter (RECIST score), percent change in tumor volume (ΔV), and difference in ratio of tumor burden to pancreas volume (ΔB) were computed as QIB for TR across the NAT interval via two approaches: image registration of the baseline tumor region of interest, and direct segmentation of the PDAC lesion in baseline and restaging images. Inter-reader agreements of RECIST, ΔV, and ΔB QIB from both approaches were compared via concordance correlation coefficient (CCC). Associations with OS and RFS were compared over 5-year follow-up via Harrell’s C-index. Results: Among expert radiologists, RECIST score CCC was 0.77 [95% CI: 0.59-0.88], whereas image registration QIB CCC values were significantly higher (ΔV: 0.95 [0.90-0.98], p=0.002; ΔB: 0.93 [0.86-0.97], p=0.014). ΔV and ΔB QIB measured by direct tumor segmentation had significantly lower CCC than by image registration (p&lt;0.001). Among novice readers, image registration QIB (ΔV: 0.83 [0.67-0.91], ΔB: 0.70 [0.48-0.84]) produced higher agreement than RECIST (0.26 [-0.17-0.61]; p=0.002 and p=0.035), with CCC more similar to expert readers. While image registration QIB did not differ from RECIST in association with RFS, registration-based QIB predicted OS with significantly higher C-index (ΔV: 0.58 [0.56-0.60], p&lt;0.001; ΔB: 0.64 [0.62-0.66], p&lt;0.001) than RECIST (0.50 [0.45-0.55]) among experts. Conclusions: Longitudinal registration of the pancreas yields more consistent QIB for PDAC TR with stronger association to OS outcomes than RECIST. These registration-based QIB aim to advance PDAC management by improving patient-specific discrimination of TR after NAT. [Table: see text]