Jesse Skoch1,2, Rizwan Tahir3, Todd Abruzzo2,4, John M Taylor5, Mario Zuccarello2, Sudhakar Vadivelu6,7,8,9. 1. Division of Neurosurgery, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH, 45229, USA. 2. Department of Neurosurgery, University of Cincinnati College of Medicine, 231 Albert Sabin Way, Cincinnati, OH, 45267, USA. 3. Department of Neurosurgery, Henry Ford Hospital, 2799 West Grand Blvd, Detroit, MI, 48202, USA. 4. Division of Radiology, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH, 45229, USA. 5. Division of Neurology, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH, 45229, USA. 6. Division of Neurosurgery, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH, 45229, USA. sudhakar.vadivelu@cchmc.org. 7. Department of Neurosurgery, University of Cincinnati College of Medicine, 231 Albert Sabin Way, Cincinnati, OH, 45267, USA. sudhakar.vadivelu@cchmc.org. 8. Division of Radiology, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH, 45229, USA. sudhakar.vadivelu@cchmc.org. 9. Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave. MLC-2016, Cincinnati, OH, 45229 - 3039, USA. sudhakar.vadivelu@cchmc.org.
Abstract
PURPOSE: Artificial neural networks (ANN) are increasingly applied to complex medical problem solving algorithms because their outcome prediction performance is superior to existing multiple regression models. ANN can successfully identify symptomatic cerebral vasospasm (SCV) in adults presenting after aneurysmal subarachnoid hemorrhage (aSAH). Although SCV is unusual in children with aSAH, the clinical consequences are severe. Consequently, reliable tools to predict patients at greatest risk for SCV may have significant value. We applied ANN modeling to a consecutive cohort of pediatric aSAH cases to assess its ability to predict SCV. METHODS: A retrospective chart review was conducted to identify patients < 21 years of age who presented with spontaneously ruptured, non-traumatic, non-mycotic, non-flow-related intracranial arterial aneurysms to our institution between January 2002 and January 2015. Demographics, clinical, radiographic, and outcome data were analyzed using an adapted ANN model using learned value nodes from the adult aneurysmal SAH dataset previously reported. The strength of the ANN prediction was measured between - 1 and 1 with - 1 representing no likelihood of SCV and 1 representing high likelihood of SCV. RESULTS: Sixteen patients met study inclusion criteria. The median age for aSAH patients was 15 years. Ten underwent surgical clipping and 6 underwent endovascular coiling for definitive treatment. One patient experienced SCV and 15 did not. The ANN applied here was able to accurately predict all 16 outcomes. The mean strength of prediction for those who did not exhibit SCV was - 0.86. The strength for the one patient who did exhibit SCV was 0.93. CONCLUSIONS: Adult-derived aneurysmal SAH value nodes can be applied to a simple AAN model to accurately predict SCV in children presenting with aSAH. Further work is needed to determine if ANN models can prospectively predict SCV in the pediatric aSAH population in toto; adapted to include mycotic, traumatic, and flow-related origins as well.
PURPOSE: Artificial neural networks (ANN) are increasingly applied to complex medical problem solving algorithms because their outcome prediction performance is superior to existing multiple regression models. ANN can successfully identify symptomatic cerebral vasospasm (SCV) in adults presenting after aneurysmal subarachnoid hemorrhage (aSAH). Although SCV is unusual in children with aSAH, the clinical consequences are severe. Consequently, reliable tools to predict patients at greatest risk for SCV may have significant value. We applied ANN modeling to a consecutive cohort of pediatric aSAH cases to assess its ability to predict SCV. METHODS: A retrospective chart review was conducted to identify patients < 21 years of age who presented with spontaneously ruptured, non-traumatic, non-mycotic, non-flow-related intracranial arterial aneurysms to our institution between January 2002 and January 2015. Demographics, clinical, radiographic, and outcome data were analyzed using an adapted ANN model using learned value nodes from the adult aneurysmalSAH dataset previously reported. The strength of the ANN prediction was measured between - 1 and 1 with - 1 representing no likelihood of SCV and 1 representing high likelihood of SCV. RESULTS: Sixteen patients met study inclusion criteria. The median age for aSAH patients was 15 years. Ten underwent surgical clipping and 6 underwent endovascular coiling for definitive treatment. One patient experienced SCV and 15 did not. The ANN applied here was able to accurately predict all 16 outcomes. The mean strength of prediction for those who did not exhibit SCV was - 0.86. The strength for the one patient who did exhibit SCV was 0.93. CONCLUSIONS: Adult-derived aneurysmalSAH value nodes can be applied to a simple AAN model to accurately predict SCV in children presenting with aSAH. Further work is needed to determine if ANN models can prospectively predict SCV in the pediatric aSAH population in toto; adapted to include mycotic, traumatic, and flow-related origins as well.
Authors: Parham Moftakhar; Daniel L Cooke; Heather J Fullerton; Nerissa U Ko; Matthew R Amans; Jared A Narvid; Christopher F Dowd; Randall T Higashida; Van V Halbach; Steven W Hetts Journal: J Neurosurg Pediatr Date: 2015-01-02 Impact factor: 2.375
Authors: S W Hetts; J Narvid; N Sanai; M T Lawton; N Gupta; H J Fullerton; C F Dowd; R T Higashida; V V Halbach Journal: AJNR Am J Neuroradiol Date: 2009-04-08 Impact factor: 3.825
Authors: G A Christoforidis; C Karakasis; Y Mohammad; L P Caragine; M Yang; A P Slivka Journal: AJNR Am J Neuroradiol Date: 2008-09-03 Impact factor: 3.825