View Presentation Video

Javad Khaghani¹, Saqib Basar¹, Yosef Chodakiewitz², Sean London², Rajpaul Attariwal¹, and Sam Hashemi^1
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Prenuvo, Vancouver, BC, Canada

Keywords: Uterus, Data Analysis

MRI is a powerful imaging technique for examining the anatomy of the female reproductive system. However, due to cost-related concerns, ease of access, acquisition time, and necessity of expert reviewers for MR images, ultrasound is the primary modality of choice. To mitigate some of these concerns, we developed an AI-driven tool comprising seven neural networks that segments the regions of interest for the whole uterus, uterine zones, ovaries, and further identifies common benign gynecological conditions. We evaluated our package on a large representative population of 2955 sagittal T2-weighted pelvic scans to obtain normative aging-curves for various regions of interest.

View Presentation Video

Yang Zhang^1,2, Liming Shi³, Xiaonan Sun³, Min-Ying Su², Taoran Cui¹, Ning J. Yue¹, and Ke Nie^1
1Department of Radiation Oncology, Rutgers-Cancer Institute of New Jersey, New Brunswick, NJ, United States, ²Department of Radiological Sciences, University of California, Irvine, CA, United States, ³Department of Radiation Oncology, Department of Radiation Oncology, Sir Run-run Shaw Hospital, Zhejiang University School of Medicine, Hangzhou, China

Keywords: Cancer, Machine Learning/Artificial Intelligence

In this study, the convolutional neural network (CNN) was implemented to segment the rectal cancer in 185 patients, with tumor ROI outlined by a radiologist on multiparametric magnetic resonance images (mp-MRI). The Dice similarity coefficient (DSC) and mean symmetrical surface distance (MSD) values were used to evaluate the results of the proposed algorithm. The mean DSC and MSD were 0.88 and 2.2 cm respectively for pre-treatment MR images. The transfer-learning model using paired pre-treatment and near-end-treatment images could also lead to acceptable result for boost volume segmentation with the mean DSC of 0.83 and MSD of 2.7 cm. Our work showed the deep-learning with combined image sequences was promising for automatic tumor localization and segmentation of locally advanced rectal cancer. After applying transfer learning with the prior information obtained from the pre-treatment images, the proposed method can be further used to identify boost volume for radiation dose escalation which has been demonstrated beneficial in improving pathological complete response rate.

View Presentation Video

Marrissa J McIntosh^1,2, Maksym Sharma^1,2, Harkiran K Kooner^1,2, Hana Serajeddini^2,3, Anurag Bhalla³, Cory Yamashita³, and Grace Parraga^1,2,3,4
1Department of Medical Biophysics, Western University, London, ON, Canada, ²Robarts Research Institute, London, ON, Canada, ³Division of Respirology, Department of Medicine, Western University, London, ON, Canada, ⁴School of Biomedical Engineering, Western University, London, ON, Canada

Keywords: Lung, Hyperpolarized MR (Gas), asthma, biologic therapy

It was previously shown that ¹²⁹Xe MR ventilation images contain embedded texture features which help predict (before treatment was initiated) those severe asthma patients who were more likely to experience early response to anti-IL-5Rα biologic therapy. Thus, we postulated that such texture features would also help identify patients with an enduring, late response. Here we identified specific ¹²⁹Xe MRI features that predicted both early and late response to anti-IL-5 therapy, which were superior to clinical measurements. These promising results suggest that ¹²⁹Xe MRI texture features sensitively predict patients with response to biologic therapy interventions.

Shasha Liu¹, Siyao DU¹, Si Gao¹, Lizhi Xie², Yuee Teng³, Feng Jin⁴, and Lina Zhang^1
1Department of Radiology, The First Hospital of China Medical University, shenyang, China, ²MR Research, GE Healthcare, Beijing, China, ³Departments of Medical Oncology and Thoracic Surgery, The First Hospital of China Medical University, shenyang, China, ⁴Department of Breast Surgery, The First Hospital of China Medical University, shenyang, China

Keywords: Breast, Radiomics, DCE-MRI

This prospective study with multiple follow-up time points investigated the early predictive value of the delta-radiomic model of axillary lymph node (ALN) using dynamic contrast-enhanced (DCE) MRI for axillary pathological complete response (pCR) in breast cancer patients after neoadjuvant chemotherapy (NAC). The results indicated that the delta-radiomic model based on early changes of ALN features performed better among all radiomic models. Moreover, when combined with clinical features, the combined model achieved the best diagnostic performance of any model we tested. The delta-radiomic + clinical model may be a promising method for ALN pCR prediction in the initial phase of NAC.

View Presentation Video

Javad Khaghani¹, Lucas Porto², Saurabh Garg¹, Saqib Basar¹, Yosef Chodakiewitz³, Sean London³, Rajpaul Attariwal¹, and Sam Hashemi^1
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Voxelwise Imaging Technology Inc., San Francisco, CA, United States, ³Prenuvo, Vancouver, BC, Canada

Keywords: Prostate, Data Analysis, Artificial Intelligence/ Machine Learning

AI-assisted prostate whole-gland and zonal volume quantification enable quantitative reproducibility and enhance read-time efficiency. Our whole-gland AI-segmentation solution enhanced measurement accuracy by 23.59%, compared with traditional volume estimates. Zonal solution enabled us to generate population normative aging-curves and we used a shallow classifier to identify patients with BPH. Our findings show the transitional zone grows 2.05 ml and 3.58 ml per decade for the entire population and patients with BPH, respectively, while peripheral zone grows 0.70 ml per decade. Further, the transitional and peripheral zones grow 2.43 and 1.4 times their size over lifetime.

View Presentation Video

Yang Zhang^1,2, Jiejie Zhou³, Zhongwei Chen³, You-Fan Zhao³, Meihao Wang³, Yan-Lin Liu², Jeon-Hor Chen², Ke Nie¹, and Min-Ying Su^2
1Department of Radiation Oncology, Rutgers-Cancer Institute of New Jersey, New Brunswick, NJ, United States, ²Department of Radiological Sciences, University of California, Irvine, CA, United States, ³Department of Radiology, First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China

Keywords: Cancer, Machine Learning/Artificial Intelligence

A deep learning model using 2D ResNet50 CNN was trained to differentiate a dataset of 103 malignant vs 73 benign breast lesions, then tested in a testing dataset of 53 malignant and 31 benign cases. The 2D slice-based results were used to calculate a probability for each lesion, by using 5 methods: (1) slice-based average, (2) tumor area weighted average, (3) tumor perimeter weighted average, (4) using the probability of the largest tumor slice, (5) using the highest probability among all slices. The results showed using the highest probability to convert from slice-based to lesion-based diagnosis had the best performance.

View Presentation Video

Elaine Yuen Phin Lee¹, Chia-Wei Li², Patricia Lan³, Xinzeng Wang⁴, Arnaud Guidon⁵, and Chien-Yuan Lin^2
1Department of Diagnostic Radiology, Queen Mary Hospital, The University of Hong Kong, Hong Kong SAR, China, ²GE Healthcare, Taipei, Taiwan, ³GE Healthcare, Menlo Park, CA, United States, ⁴GE Healthcare, Houston, TX, United States, ⁵GE Healthcare, Boston, MA, United States

Keywords: Pelvis, Machine Learning/Artificial Intelligence, Deep learning reconstruction, Multi-shot DWI

This study introduced the deep learning reconstruction (DLRecon) to multi-shot DWI (MUSE-DWI) in the pelvis and aimed to investigate the changes in SNR and image quality with MUSE-DWI DLRecon. Compared with the MUSE-DWI non-DLRecon, the MUSE-DWI DLRecon showed higher SNR with stable ADC quantification. With that, DLRecon could reduce image distortion using higher shots MUSE-DWI with comparable SNR and scan time to clinically used 2-shot MUSE-DWI. This preliminary result showed the potential power of DLRecon in the pelvis allowing higher shots MUSE-DWI to be integrated in clinical practice to reduce image distortion.

View Presentation Video

Stephane Loubrie¹, Ana Rodriguez-Soto¹, Tyler Seibert^1,2,3, Michael Hahn¹, Vandana Dialani^4,5, Catherine Wei⁶, Zahra Karimi⁴, Joshua Kuperman¹, Anders Dale^1,7, Etta Pisano^4,8, Jingjing Zou⁹, and Rebecca Rakow-Penner^1,3
1Radiology, University of California, San Diego, San Diego, CA, United States, ²Radiation medicine, University of California, San Diego, San Diego, CA, United States, ³Bioengineering, University of California, San Diego, San Diego, CA, United States, ⁴Radiology, Beth Israel Hospital, Boston, MA, United States, ⁵Harvard Medical School, Boston, MA, United States, ⁶Commonwealth Radiology Associates, Boston, MA, United States, ⁷Neurosciences, University of California, San Diego, San Diego, CA, United States, ⁸American College of Radiology, Reston, VA, United States, ⁹Biostatistics, Herbert Wertheim School of Public Health and Human Longevity Science, University of California, San Diego, San Diego, CA, United States

Keywords: Breast, Diffusion/other diffusion imaging techniques

Diffusion weighted imaging (DWI) holds great potential in improving specificity of findings detected on contrast enhanced breast MRI. Restriction spectrum imaging (RSI), an advanced diffusion imaging model, has potential in discriminating between malignant and fibroglandular breast tissue. In this abstract, we evaluate RSI’s performance, combined with a random forest model, in differentiating lesions requiring biopsies from lesions that do not. The model showed significant preliminary results.

View Presentation Video

Yanyan Xie¹, shangxuan Li¹, Baoer Liu², Yikai Xu², and Wu Zhou^1
1School of Medical Information Engineering, Guangzhou University of Chinese Medicine, Guangzhou, China, ²Department of Medical Imaging Center, Nanfang Hospital, Southern Medical University, Guangzhou, China

Keywords: Liver, Machine Learning/Artificial Intelligence

Multi-task learning has been widely used for jointly tumor segmentation and classification. Uncertainty estimation of the subtask weight coefficient in multi-task learning has been investigated. However, due to the presence of noise in medical image, data uncertainty will affect the performance of multi-task learning. In addition, model uncertainty has not been conducted for multi-task learning. In this work, we propose a triplet-uncertainty in multi-task deep learning network (TU-MTL), simultaneously considering the uncertainty estimation of subtask weight coefficient, data uncertainty estimation and model uncertainty estimation. Experimental results of clinical hepatocellular carcinoma (HCC) demonstrate the effectiveness of the proposed method.

View Presentation Video

Lingsong Meng¹, Xin Zhao¹, Jinxia Guo², Lin Lu¹, Meiying Cheng¹, Qingna Xing¹, Honglei Shang¹, Yan Chen¹, Penghua Zhang¹, and Xiaoan Zhang^1
1The Third Affiliated Hospital of Zhengzhou University, Zhengzhou, China, ²General Electric (GE) Healthcare, MR Research China, Beijing, Beijing, China

Keywords: Breast, Radiomics

Improving the assessment of Breast Imaging Reporting and Data System (BI-RADS) 4 lesions can avoid unnecessary biopsies. As an emerging field, radiomics has been successfully explored as a means to aid decision-making for the diagnosis and risk stratification of several kinds of cancers^1-4. In this study, we combined radiomics features extracted from ultrafast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) (using the Differential sub-sampling with cartesian ordering (DISCO) technique) with an artificial neural network (ANN) to improve diagnostic performance in assessing BI-RADS 4 lesions and evaluate the potential to avoid unnecessary biopsies.

View Presentation Video

Wanli Zhang^1,2, Fangrong Liang^1,2, Jiamin Li^1,2, Yongzhou Xu³, Aaron Zhang³, Xinqing Jiang^1,2, Xin Zhen⁴, and Ruimeng Yang^1,2
1Department of Radiology, The Second Affifiliated Hospital, School of Medicine, South China University of Technology, Guangzhou, China, ²Department of Radiology, Guangzhou First People’s Hospital, Guangzhou, China, ³Philips Healthcare, Guangzhou, China, ⁴School of Biomedical Engineering, Southern Medical University, Guangzhou, China

Keywords: Breast, Radiomics, breast cancer, molecular receptor status, feature fusion

Since breast cancer is a highly heterogeneous tumor, prognosis and treatment response differ significantly according to different molecular subtypes. Based on multiparametric magnetic resonance imaging (MRI), we developed a feature fusion radiomics (R_FF) model and investigated its performance in identifying the molecular receptor status of breast cancer. Mimicking the diagnostic approach of the radiologists by integrating image information from different MR sequences, the R_FF model outperformed any single MRI-based radiomics model, demonstrating its potential for molecular subtypes classification of breast tumors.

View Presentation Video

Anum Masood^1,2, Sølvi Knapstad², Håkon Johansen³, Trine Husby³, Live Eikenes ¹, Mattijs Elschot ^1,3, and Pål Erik Goa^2,3
1Department of Circulation and Medical Imaging, NTNU, Trondheim, Norway, ²Department of Physics, NTNU, Trondheim, Norway, ³Department of Radiology and Nuclear Medicine, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway

Keywords: Machine Learning/Artificial Intelligence, PET/MR, Automated Segmentation, Deep Learning, Transformers

Automated segmentation of tumors and designing a computer-assisted diagnosis system requires co-learning of imaging features from the complementary PET/MRI. We aim to develop an automated method for the segmentation of cancer-affected lymph nodes on Positron Emission Tomography/Magnetic Resonance Imaging (PET/MRI) using a modified Swin Transformer model (ST) integrated with a novel Multi-Scale Feature Fusion & Reorganization Module (MSFFRM). Our results with Hodgkin lymphoma (HL) and Non-Hodgkin lymphoma subtype Diffuse Large B-Cell Lymphoma (DLBCL) datasets show that our proposed model ST-MSFFRM achieved better performance in lesion segmentation in comparison to other state-of-the-art segmentation methods. 

Lili Wang¹, Junqiang Lei¹, Junfeng Li², Shunlin Guo², Gang Wang², and Rui Wang^3
1Radiology, First Hospital of Lanzhou University, Lanzhou, China, ²First Hospital of Lanzhou University, Lanzhou, China, ³First Clinical Medical School of Lanzhou University, Lanzhou, Chile

Keywords: Liver, Radiomics

To preoperatively predict MVI in HCC patients, this study developed and validated an MVI nomogram prediction model that reveals the features derived from tumor and peritumor tissue of different sequences in Gd-EOB-DTPA dynamic contrast-enhanced MR and combines the clinical and radiological signatures. The nomogram included AFP, capsule appearance, arterial peritumoral enhancement, RVI, and radiomics score, which were independent risk factors for MVI. Finally, this nomogram model achieved satisfactory performance in predicting MVI in both the training and validation cohorts. Moreover, the RFS of the nomogram model was similar to the histopathology outcome.

Zhijun Geng¹, Yunfei Zhang², Yongming Dai², and Chuanmiao Xie^1
1Sun Yat-sen University Cancer Center, Guangzhou, China, ²MR Collaboration, Central Research Institute, United Imaging Healthcare, Shanghai, China

Keywords: Liver, Radiomics, Micro-Vascular Invasion

MVI has widely been considered as an important prognostic biomarker of hepatocellular carcinoma (HCC). This study aims to non-invasively identify MVI in HCC with susceptibility-weighted-imaging (SWI)-derived radiomics. The results indicated that SWI-derived radiomics are valuable for noninvasively and accurately identifying the MVI status of HCC. Furthermore, the integration of radiomics and clinical factors yielded a predictive nomogram with satisfying diagnostic performance and potential clinical benefits.

Zhehan Shen¹, Ruokun Li¹, and Fuhua Yan^1
1Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University, Shanghai, China

Keywords: Liver, Radiomics, drug-induced liver injury

Drug-induced liver injury (DILI) occurs in a small fraction of individuals exposed to drugs, herbs or dietary supplements. DILI is challenging because of the lack of biomarkers to predict chronic outcomes. In our study, we developed a radiomics model with excellent performance for prediction of chronic outcome based on MRI and clinical data. Decision curve analysis confirmed the clinical utility of the radiomics model. This model may be used to stratify DILI patients at high risk of chronic outcome.

View Presentation Video

Diana Bencikova¹, Sarah Poetter-Lang¹, Marcus Raudner¹, Martin Krššák², Siegfried Trattnig^1,3, and Ahmed Ba-Ssalamah^1
1Department of Biomedical Imaging and Image-Guided Therapy, Medical University Vienna, Vienna, Austria, ²Division of Endocrinology and Metabolism, Department of Medicine III, Medical University Vienna, Vienna, Austria, ³Christian Doppler Laboratory for Clinical Molecular Imaging, Vienna, Austria

Keywords: Liver, Machine Learning/Artificial Intelligence, Elastography, MR Value, Radiomics, Relaxometry

MR elastography is currently the most accurate non-invasive diagnostic method for liver fibrosis. However, other pathologic processes co-existing with liver fibrosis influence the stiffness measurement and including other MR-based measures might improve fibrosis assessment. In this work we suggest to add texture analysis features calculated on liver T2 maps together with T2 values from fast radial turbo-spin-echo sequence into a machine-learning classification model and to compare the performance of the model after adding selected parameters. Our results show that including both, texture analysis and T2 values significantly improves the classification performance of the model.

View Presentation Video

Bai min^1
1Radiology, Liaocheng People's Hospital, Liao'cheng, China

Keywords: Uterus, Radiomics

Preoperative risk stratification of endometrial cancer (EC) is one of the important prognostic factors and is crucial for clinical treatment planning. We aimed to evaluate the predictive value of radiomics features based on multiparameter magnetic resonance imaging (MP-MRI) for preoperative risk stratification in patients with EC. The clinical and radiomic combined predictive models has a better performance than the model based on clinical characteristics. In the training set and the test set, the IDI and he reclassification improvement index of combined predictive models is higher than that of clinical model.

View Presentation Video

Jin Fang¹, Xiao Zhang², Xiaoyun Liang², Feng Huang², and Shuixing Zhang^1
1The First Affiliated Hospital of Jinan University, Guangzhou, China, ²Neusoft Medical Systems Co, Shanghai, China

Keywords: Uterus, Radiomics

Appropriate management and treatment decisions for cervical cancer depend on accurate staging, but the differentiation between IIA and IIB by imaging assessment remains difficult. This retrospective study investigated the performance of intratumoral and peritumoral multi-parameter MRI texture information in differentiating IIA and IIB using non-invasive radiomics analysis, which was also compared with clinical factors and imaging assessment by radiologists. Among all the comparisons,  peritumoral-based radiomics models outperformed the radiologists and performed the best. This study offers a viable approach in non-invasively and accurately differentiating IIA and IIB for cervical cancer based on peritumoral texture information.

View Presentation Video

Alexander J Daniel¹, Eleanor F Cox¹, Rachael A Evans², Louise V Wain², Christopher Brightling², Elizabeth Tunnicliffe^3,4, Stefan Neubauer^3,4, Betty Raman^3,4, and Susan T Francis^1
1Sir Peter Mansfield Imaging Centre, University of Nottingham, Nottingham, United Kingdom, ²Institute for Lung Health, Leicester NIHR Biomedical Research Centre, University of Leicester, Leicester, United Kingdom, ³Division of Cardiovascular Medicine, Radcliffe Department of Medicine, Oxford University, Oxford, United Kingdom, ⁴Oxford University Hospitals NHS Foundation Trust, Oxford University, Oxford, United Kingdom

Keywords: Kidney, Segmentation

Measures of total kidney volume (TKV) help to evaluate disease progression, and masks to define the kidney are important for the automatic assessment of multiparametric images collected in the same data space. For accurate measures in multicentre studies an automated method which is vendor agnostic and robust against image artefacts is needed.  Here a single-vendor convolutional neural network is retrained and shown to be accurate on two vendors of scanner and robust against image artefacts associated with wrapping in the phase-encode direction.

View Presentation Video

Junyu Guo¹, Keith Hulsey¹, Yin Xi¹, and Ivan Pedrosa^1
1Radiology, UT Southwestern Medical Center, Dallas, TX, United States

Keywords: Kidney, Machine Learning/Artificial Intelligence

Deep learning has been successful in predicting tumor malignancy. Clear cell renal carcinoma (ccRCC) diagnosis may help in decision making between active surveillance and definitive intervention. In this study, we investigate the effects of different factors on the performance of deep learning classification of ccRCC using T2w images. We demonstrate that the performance of 15 different CNN models varied substantially.  The choice of CNN models, the cropped image size, and the type of inputs greatly altered the performance of ccRCC classification. We achieved the best AUC value of 0.81, which is close to the reported performance of radiologists.