The Predictive Value of AI Image Regression Analysis of Rhinograms for Patient Symptoms: More Than Resistance Testing
Dr Rhea Darbari Kaul, ENT Registrar, Macquarie University Hospital, Sydney, Australia
Authors List
Darbari Kaul, R., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University, Sydney, Australia
Choy, C., Hua, E., Grouse, L., Haghighi, M., Liang, K., Thiel, C., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales, Sydney, Australia
Azemi, G., Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Liu, S., Centre for Health Informatics, Australian Institute of Health Innovation, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Sacks, R., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Faculty of Medicine, University of Sydney, Australia.
Campbell, RG., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Department of Otolaryngology Head and Neck Surgery, Royal Prince Alfred Hospital, Sydney, Australia
Kalish, LH., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Faculty of Medicine, University of Sydney; Department of Otolaryngology, Head and Neck Surgery, Concord General Hospital, University of Sydney, Australia
Di Ieva, A., Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Harvey, RJ., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; School of Clinical Medicine, St Vincent's Healthcare Clinical Campus, Faculty of Medicine and Health, UNSW Sydney, Australia
Introduction: Rhinomanometry, a reference measure for the nasal airway, is often considered a research tool as resistance only weak-to-moderately correlates with patient symptoms. However, like lung spirometry curves (spirograms) offer information well beyond FEV, the rhinomanometry curve (rhinogram) has characteristics beyond simple nasal resistance. This study applied artificial intelligence (AI) to develop a predictive model examining the link between rhinograms and patient-reported outcomes (PROMs).
Methods: A diagnostic cross-sectional study was conducted on patients from a rhinology clinic. Rhinomanometry curves were assessed via image regression with convolutional neural networks (CNN) on a visual AI platform (Ximilar). PROMs included nasal obstruction (ordinal/VAS), the nasal subdomain and total SNOT-22. There was an 80/20 training/testing split. The primary outcome was a correlation coefficient (R2 score). Prediction errors were assessed with mean absolute error (MAE) and root mean squared error (RMSE). Heatmaps were generated for regions of interest in CNN decision making.
Results: Two hundred patients (age 44±17yrs, 48% female) were analysed. Analysis of rhinograms demonstrated high to strong correlation (0.7-0.9) with R2 scores of SNOT-22(0.84), nasal subdomain(0.79), nasal obstruction (ordinal 0.77/VAS 0.75). The prediction errors (MAE/RMSE) were; SNOT22 (24/30), nasal subdomain (8/9), nasal obstruction ordinal (2/2) and VAS (22/27).
Conclusion: This study highlights the strong relationship between nasal airflow analysis and patient reported outcomes. It is highly likely that the rhinogram produced from rhinomanometry may contain significantly more value to clinical care when assessed beyond simple resistance measures at 150Pa.
Darbari Kaul, R., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University, Sydney, Australia
Choy, C., Hua, E., Grouse, L., Haghighi, M., Liang, K., Thiel, C., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales, Sydney, Australia
Azemi, G., Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Liu, S., Centre for Health Informatics, Australian Institute of Health Innovation, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Sacks, R., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Faculty of Medicine, University of Sydney, Australia.
Campbell, RG., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Department of Otolaryngology Head and Neck Surgery, Royal Prince Alfred Hospital, Sydney, Australia
Kalish, LH., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Faculty of Medicine, University of Sydney; Department of Otolaryngology, Head and Neck Surgery, Concord General Hospital, University of Sydney, Australia
Di Ieva, A., Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Human and Health Sciences, Macquarie University, Sydney, Australia
Harvey, RJ., Rhinology and Skull Base Research Group, Applied Medical Research Centre, University of New South Wales; Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University; School of Clinical Medicine, St Vincent's Healthcare Clinical Campus, Faculty of Medicine and Health, UNSW Sydney, Australia
Introduction: Rhinomanometry, a reference measure for the nasal airway, is often considered a research tool as resistance only weak-to-moderately correlates with patient symptoms. However, like lung spirometry curves (spirograms) offer information well beyond FEV, the rhinomanometry curve (rhinogram) has characteristics beyond simple nasal resistance. This study applied artificial intelligence (AI) to develop a predictive model examining the link between rhinograms and patient-reported outcomes (PROMs).
Methods: A diagnostic cross-sectional study was conducted on patients from a rhinology clinic. Rhinomanometry curves were assessed via image regression with convolutional neural networks (CNN) on a visual AI platform (Ximilar). PROMs included nasal obstruction (ordinal/VAS), the nasal subdomain and total SNOT-22. There was an 80/20 training/testing split. The primary outcome was a correlation coefficient (R2 score). Prediction errors were assessed with mean absolute error (MAE) and root mean squared error (RMSE). Heatmaps were generated for regions of interest in CNN decision making.
Results: Two hundred patients (age 44±17yrs, 48% female) were analysed. Analysis of rhinograms demonstrated high to strong correlation (0.7-0.9) with R2 scores of SNOT-22(0.84), nasal subdomain(0.79), nasal obstruction (ordinal 0.77/VAS 0.75). The prediction errors (MAE/RMSE) were; SNOT22 (24/30), nasal subdomain (8/9), nasal obstruction ordinal (2/2) and VAS (22/27).
Conclusion: This study highlights the strong relationship between nasal airflow analysis and patient reported outcomes. It is highly likely that the rhinogram produced from rhinomanometry may contain significantly more value to clinical care when assessed beyond simple resistance measures at 150Pa.
