|
Researcher uses WSG neural networks in the study and treatment of rheumatoid arthritis (RA).
Long time user Guirish Solanki is a researcher at the Dept of Surgical Neurology, The National Hospital, Queen Square, London. Since the mid-nineties he and his colleagues have published several impressive papers describing use of our neural networks in the study and treatment of rheumatoid arthritis (RA).
Solanki describes his work in his own words:
"Beginning with the ability to differentiate a non-rheumatoid patient from a rheumatoid patient, hitherto requiring at least 6 months for confirmation, now just with 4 basic inputs the ANN can tell you the answer! We then went further to see if the ANN could predict at which stage the patient was in the spectrum of disease. In this regard, particularly, was it possible to predict when a patient would require surgery ?"
"By combining 2 large databases of evidence one containing rheumatoid patients in their early stage and another consisting of severely advanced disease who had undergone surgery, we pinpointed just a handful of inputs that determined which patients were in their early stages (non-surgical) and those that were surgical cases. This study has profound logistical and outcome modifying implications when dealing with patients with this dreadful disease. If surgery is delayed too long (the patient is bed-bound) surgery is ineffective and mean survivalis in months. Yet if surgery is carried out at an earlier time there is a more than reasonable chance of several years of disease-controlled survival."
Here are a few of Solanki's papers:
Glasgow Outcome Scale following Aneurysmal Subarachnoid Haemorrhage: Improved prediction using a neural network with genetic optimisation
G. Solanki, J. Grieve, H. Ellamushi, T. Paleologos, S. Beales, N. Kitchen
Society of British Neurological Surgeons, 31 August -3 September 1999, Cork.
Artificial Intelligence Neural Networks: Diagnosis and Outcome Prediction in Rheumatoid Cervical Disease
G. Solanki, T. Watson, A. Singh, A. Young, H.A. Crockard
BRITSPINE 99 Meeting, 3 - 5th March 1999, Manchester
Artificial Neural Networks: Diagnostic Prediction in Rheumatoid Arthritis of the Cervical Spine.
Solanki G, H A Crockard. Presented at the British Cervical Spine Research Society Meeting, The Copthorne Hotel, Newcastle-upon-Tyne, 8-9/11/96
Diagnostic Prediction in Rheumatoid Arthritis using Neuronal Net Models.
Solanki G.
Presented at the Postgraduate Research Meeting, Institute of Neurology, University of London, 19 September 1996.
Artificial Intelligence: Can It Identify The Rheumatoid Neck ? Solanki G, Watson T, Singh A, Tammam A, Young A, Crockard H A.
British Journal of Neurosurgery Vol. II (5), pp 481, 1997
Artificial Intelligence Neural Networks: Diagnosis and Outcome Prediction in Rheumatoid Cervical Disease
G. Solanki, H Alan Crockard.
Proceedings of BRITSPINE 99 Meeting, 3 - 5th March 1999, Manchester (In Print JBJS (UK))
Glasgow Outcome Scale following Aneurysmal Subarachnoid Haemorrhage: Improved prediction using a neural network with genetic optimisation
G. Solanki, J. Grieve, H. Ellamushi, T. Paleologos, S. Beales, N. Kitchen
Proceedings of the Society of British Neurological Surgeons, 31 August -3 September 1999, Cork, (In print British Journal of Neurosurgery)
We present here a synopsis of one of one paper by G SOLANKI and HA CROCKARD:
Dept of Surgical Neurology, The National Hospital, Queen Square, London WC1N 3BG
ARTIFICIAL INTELLIGENCE: PREDICTING POST-SURGICAL FUNCTIONAL OUTCOME IN THE ADVANCED RHEUMATOID NECK.
INTRODUCTION: Cervical spine involvement is the second commonest manifestation of Rheumatoid arthritis (RA) and potentially the most serious. Cervical spine instability causes repetitive concussive microtrauma and eventually spinal cord injury. Neurological deterioration ensues and if it is such that the patient is bed-bound, then surgery has been shown to be largely ineffective. Yet, while a pre-emptive surgical strike may be appealing, surgery is not without considerable risk, and the patient may be exposed to unnecessary risk if operated sooner than it should have been. The ideal timing for rheumatoid neck surgery, thus, is a source of considerable debate.
AIMS: Using artificial intelligence neural networks (ANN) technology, develop models that predict the pre and post-surgical functional state and or outcome thereby providing decision support, to determine the optimum time for surgery. By reverse engineering these networks, identify which clinically relevant factors have contributed to the network's prediction.
MATERIALS & METHODS: Overall 194 surgical rheumatoid cases (37 males and 157 females, aged 13-82, RA 5 to 60 years duration) recruited from the prospective database maintained over 10 years by the senior author and 435 non-surgical RA patients thus far recruited from the ERAS - Early Rheumatoid Arthritis Study - prospective observational database (152 males, 283 females, aged 17-83, RA duration 0 to 10 years) were the subjects of this study. Pre and post operative (in the surgical group) demographic, clinical, radiological and functional profile data was collected. Measurements were made on lateral flexion and extension cervical radiographs. About 90% of the above data profiles were used to train and internally test the 3-layer feedforward with backpropagation, and general regression with genetic optimisation neural models. These were then externally validated on a new set of data not previously exposed to the neural networks (about 10% of the data set). Further validation was obtained using more conventional tests of statistical significance. The validated neural architectures have been programmed into operational neural networks that can be 'fired' from applications such as Microsoft Excel, or clinical workstations such as the Reports On Call by the use of dynamic link libraries.
RESULTS: RA Progression Model: Correctly predicted a patient as a surgical case or as a non-surgical case 98% of the time, Spearman correlation = 0.92, r2= 0.85. Pre-Operative Functional Status Model: 88% correct prediction of pre-operative HAQ scores. r =0.88, r2 =0.71. Post-Operative Functional Outcome Model: 80 % Correct prediction of Post-operative HAQ scores r=0.89, r2=0.79.
Reverse engineering of the neural networks showed that the models had applied significance to the inputs similar to those used by clinicians (age, sex, severity of vertical translocation, disease duration and functional state). Models were validated with data not previously exposed to them.
CONCLUSION:
1. Accurate prognostic prediction in rheumatoid cervical spine disease is possible using artificial neural networks.
2. Reverse Net engineering offers insight into clinically significant prognostic factors.
We feel this is an encouraging basis towards predicting the ideal timing of surgery. |
