Tommaso Di Noto¹, Chirine Atat¹, Eduardo Gamito Teiga¹, Monika Hegi², Andreas Hottinger³, Patric Hagmann¹, Meritxell Bach Cuadra¹, and Jonas Richiardi^1
1Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ²Neuroscience Research Center, CHUV, Lausanne, Switzerland, ³Department of Clinical Neurosciences, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland

Manual annotations are a major bottleneck in supervised machine learning. We present a method that leverages Natural Language Processing (NLP) to generate automatic weak labels from radiology reports. We show how weak labels can be used for the image classification task of high-grade-glioma diagnostic surveillance. We apply a convolutional neural network (CNN) to classify T2w difference maps that either indicate tumor stability or instability. Results suggest that pretraining the CNN with weak labels and fine-tuning it on manually-annotated data leads to better performance (though not statistically significant) when compared to a baseline pipeline where only manually annotated data is used. 

Ya Zhang¹, Dehong Luo², Wei Guo³, Zhou Liu², Dan Bao¹, Haijun Xu¹, Lin Li¹, Meng Lin¹, Yanfeng Zhao¹, and Xinming Zhao^1
1Radiology, National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medica, Beijing, China, ²Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Shenzhen, China, ³Radiology, Peking University Third Hospital, Beijing, China

To investigate the predicted survival risk power of three IVIM-DWI models in laryngeal and hypopharyngeal squamous cell carcinoma (LHSCC) patients after (chemo)radiotherapy, 45 patients were retrospectively enrolled. IVIM-DWI sequence scanning was performed using echo planner imaging (EPI) sequence with 12 b values (0, 10, 20, 30, 50, 70, 100, 150, 200, 400, 800 and 1000 s/mm2). Followed up for 2-72 months with median follow-up of 14 months. The pretreatment ADC, D, f value and D* value were significantly correlated with prognosis of local disease, and the pretreatment ADC, D* value were independent biomarkers for survival risk prediction.

Eric Aliotta¹, Yu-Chi Hu¹, Peng Zhang¹, Phil Lichtenwalner¹, Amanda Caringi¹, Natasha Allgood¹, Jillian Tsai², Kaveh Zakeri², PengPeng Zhang¹, Nancy Lee², Laura Cervino¹, and Michalis Aristophanous^1
1Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, ²Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, NY, United States

Longitudinal MRI is an ideal tool for tracking anatomic changes that occur during head and neck radiotherapy. We have implemented an image tracking system that monitors volumetric changes in gross tumor volumes (GTV) and parotid glands to identify and alert of major changes early in treatment. In a cohort of 91 patients, this system identified systematic shrinkage of GTVs (9.2±8.3% per week) and parotids (3.1±3.7% per week) during treatment. Importantly, GTV changes observed in the first week of treatment were strongly predictive of larger changes that would occur later in the course (P<1x10^-5, two-tailed t-test). 

Jinghong Liu¹ and Ailian Liu^1
1The first hospital of Dalian medical university, Dalian, China

Clear cell carcinoma of kidney usually presents with high signal intensity on T2WI imaging, while some clear cell carcinoma presents with low signal intensity on T2WI imaging.Other rare renal tumors, such as papillary renal cell carcinoma and lipid-deficient angiomyolipoma, are characterized by low signal intensity on T2WI.  Intravoxel incoherent motion (IVIM) is a model built on the basis of DWI, reflecting the structural changes and physiological state of cells from the microscopic level. Our objectivie si to evaluate the feasibility of the IVIM in differentiating renal tumors with low signal intensities on T2WI.

Xiaoling Gong¹, Yu Shen², Daguang Wen³, Mingtian Wei², Xiaoxiao Zhang⁴, Xiaoyong Zhang⁴, Ziqiang Wang², and Bing Wu^3
1Departments of Radiology, West China Hospital, Sichuan University, Cheng du, China, ²Department of Clinical, West China Hospital, Sichuan University, Chengdu, China, ³Departments of Radiology, West China Hospital, Sichuan University, Chengdu, China, ⁴Philips Healthcare, China, Chengdu, China

Neoadjuvant chemotherapy can improve oncologic outcomes and overall survival rate. However, not all patients respond well during treatment, and continuous treatment will be painful, costly, and even increase risk of progression. Thus, early stratification of treatment response and screening of patients who may not be effective in treatment is very important. We investigated the value of T1 and T2 mapping MRI in response evaluation and found that T2 value showed a significant difference between groups of pre and 1-week post-treatment, while the T1 value had no statistical difference. T2 mapping may be a useful functional biomarker for early responses evaluation.