Machine Learning Applications in Head and Neck Radiation Oncology: Lessons From Open-Source Radiomics Challenges.

Radiomics leverages existing image datasets to provide non-visible data extraction via image post-processing, with the aim of identifying prognostic, and predictive imaging features at a sub-region of interest level. However, the application of radiomics is hampered by several challenges such as lack of image acquisition/analysis method standardization, impeding generalizability. As of yet, radiomics remains intriguing, but not clinically validated. We aimed to test the feasibility of a non-custom-constructed platform for disseminating existing large, standardized databases across institutions for promoting radiomics studies. Hence, University of Texas MD Anderson Cancer Center organized two public radiomics challenges in head and neck radiation oncology domain. This was done in conjunction with MICCAI 2016 satellite symposium using Kaggle-in-Class, a machine-learning and predictive analytics platform. We drew on clinical data matched to radiomics data derived from diagnostic contrast-enhanced computed tomography (CECT) images in a dataset of 315 patients with oropharyngeal cancer. Contestants were tasked to develop models for (i) classifying patients according to their human papillomavirus status, or (ii) predicting local tumor recurrence, following radiotherapy. Data were split into training, and test sets. Seventeen teams from various professional domains participated in one or both of the challenges. This review paper was based on the contestants' feedback; provided by 8 contestants only (47%). Six contestants (75%) incorporated extracted radiomics features into their predictive model building, either alone (n = 5; 62.5%), as was the case with the winner of the "HPV" challenge, or in conjunction with matched clinical attributes (n = 2; 25%). Only 23% of contestants, notably, including the winner of the "local recurrence" challenge, built their model relying solely on clinical data. In addition to the value of the integration of machine learning into clinical decision-making, our experience sheds light on challenges in sharing and directing existing datasets toward clinical applications of radiomics, including hyper-dimensionality of the clinical/imaging data attributes. Our experience may help guide researchers to create a framework for sharing and reuse of already published data that we believe will ultimately accelerate the pace of clinical applications of radiomics; both in challenge or clinical settings.

A comparison between gadofosveset trisodium and gadobenate dimeglumine for steady state MRA of the thoracic vasculature.

PURPOSE: Retrospective comparison between gadofosveset trisodium and gadobenate dimeglumine steady state magnetic resonance angiography (SS-MRA) of the thoracic vasculature at 1.5T using signal-to-noise ratio (SNR) and vessel edge sharpness (ES) as markers of image quality. MATERIALS AND METHODS: IRB approval was obtained. Twenty separate patients each underwent SS-MRA using high-resolution 3D ECG-triggered coronal IR-TFE at 1.5T approximately 3-4 minutes following 10 or 15 mL gadofosveset or 20 mL gadobenate. ROIs were placed in the right atrium, left ventricle, left atrium, ascending aorta, descending aorta, and right pulmonary artery to estimate SNR. Vessel ES was estimated as 20-80% rise distances from line intensity profiles in the left pulmonary vein, ascending aorta, and descending aorta. Data were analyzed using nonpaired Student's t-test (threshold for significance set at P < 0.05). RESULTS: There was no significant difference in mean SNR for the gadofosveset or gadobenate groups (P values: 0.14 to 0.85). There was no significant difference in mean vessel ES for gadofosveset and gadobenate groups (P values: 0.17 to 0.78). CONCLUSION: High quality thoracic SS-MRA can be achieved with gadobenate dimeglumine, similar to that achieved with the blood pool agent gadofosveset trisodium provided that imaging is initiated quickly (3-4 min) after contrast injection.

Multisite near-infrared spectroscopy predicts elevated blood lactate level in children after cardiac surgery.

OBJECTIVES: To determine if a relationship exists between regional oxyhemoglobin saturation (rSO(2)) measured at various body locations by near-infrared spectroscopy (NIRS) and blood lactate level in children after cardiac surgery. DESIGN: A prospective, observational study. SETTING: A pediatric cardiac intensive care unit in a university hospital. PARTICIPANTS: Twenty-three children undergoing repair of congenital heart disease. Patients with single-ventricle physiology and/or residual intracardiac shunts were excluded. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Cerebral, splanchnic, renal, and muscle rSO(2) values were recorded every 30 seconds via NIRS for 24 hours postoperatively. Blood lactate levels measured minimally at 0, 2, 4, 6 and 24 hours postoperatively were correlated with rSO(2) values derived by averaging all values recorded during the 60 minutes preceding the blood draw. Twenty-three patients were enrolled with 163 lactate measurements and more than 39,000 rSO(2) observations analyzed. Cerebral rSO(2) had the strongest inverse correlation with lactate level followed by splanchnic, renal, and muscle rSO(2) (r = -0.74, p < 0.0001, r = -0.61, p < 0.0001, r = -0.57, p < 0.0001, and r = -0.48, p < 0.0001, respectively). The correlation improved by averaging the cerebral and renal rSO(2) values (r = -0.82, p < 0.0001). Furthermore, an averaged cerebral and renal rSO(2) value or=3.0 mmol/L with a sensitivity of 95% and a specificity of 83% (p = 0.0001). CONCLUSIONS: Averaged cerebral and renal rSO(2) less than 65% as measured by NIRS predicts hyperlactatemia (>3 mmol/L) in acyanotic children after congenital heart surgery. Hence, this noninvasive, continuous monitoring tool may facilitate the identification of global hypoperfusion caused by low cardiac output syndrome in this population.