diff --git a/README.md b/README.md
index af08d03..a7ad682 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,20 @@
-# NiiVue brainchop
+### NiiVue brainchop
 
 Provided a lesion map and an acute [Center of Cancellation](https://github.com/neurolabusc/Cancel) score, this web page will predict the chronic neglect severity. The provided lesion map should be normalized to standard space. This is an edge-based solution: the values are computed directly on the web browser, and are not shared with a cloud service. [Try the live demo](https://niivue.github.io/niivue-neglect/)
 
-### To serve locally
+### Usage
+
+ 1. Open the [live demo](https://niivue.github.io/niivue-neglect/).
+ 2. Press the `Open Lesion Map` button and select the normalized lesion. You can normalize lesions using [the clinical toolbox for SPM](https://github.com/neurolabusc/Clinical).
+ 3. You can visually inspect the lesion (shown as red) with respect to the grayscale atlas and the core prediction voxels (shown in blue).
+ 4. Set the [Center of Cancellation (CoC)](https://github.com/neurolabusc/Cancel) score that was measured acutely.
+ 5. Press the `Prediction` button to see the expected outcome score.
+
+![niivue-neglect user interface](niivue-neglect.png)
+
+### For Developers
+
+You can serve a hot-reloadable web page that allows you to interactively modify the source code.
 
 ```bash
 git clone https://github.com/niivue/niivue-neglect
diff --git a/index.html b/index.html
index a3d1373..f4df563 100644
--- a/index.html
+++ b/index.html
@@ -12,7 +12,7 @@
   <header>
     <button id="openBtn">Open Lesion Map</button>
     <label for="cocNumber">Acute CoC Score</label>
-    <input style="width: 7em" type="number" id="cocNumber" min="-1" value ="0.91" max="1" />
+    <input style="width: 7em" type="number" id="cocNumber" min="-1" value ="0.65" max="1" />
     <button id="predictBtn">Prediction</button>
     <label for="maskSlider">Mask Opacity</label>
     <input
diff --git a/main.js b/main.js
index 3caf5c4..844f82e 100644
--- a/main.js
+++ b/main.js
@@ -7,9 +7,9 @@ async function main() {
   }
   function norm0to1(val, mn, mx) {
     // return clamped/normalized 0..1, linearly interpolated min..max
-    let range = mx - mn;
-    let ret = (val - mn)/range;
-    return Math.min(Math.max(ret,0),1);
+    let range = mx - mn
+    let ret = (val - mn)/range
+    return Math.min(Math.max(ret,0),1)
   }
   function neglect_predict() {
     if (nv1.volumes.length !== 3) {
@@ -60,9 +60,46 @@ async function main() {
     let acuteCoC = parseFloat(cocNumber.value)
     acuteCoC = norm0to1(acuteCoC, -0.024243014, 0.951938077)
     let ROI_volML = lesionVol / 1000
-    ROI_volML = norm0to1(ROI_volML, 0, 21.625);
-    const input_vector = [PC[0], PC[1], PC[2], PC[3], PC[4], acuteCoC, ROI_volML];
+    ROI_volML = norm0to1(ROI_volML, 0, 21.625)
+    const input_vector = [PC[0], PC[1], PC[2], PC[3], PC[4], acuteCoC, ROI_volML]
     console.log(input_vector)
+    let prediction_sum = 0
+    for (let m = 0; m < models.length; m++) {
+      const model = models[m]
+      // Extract information from the current model
+      let support_vectors_i = model.SVs
+      let coefficients_i = model.sv_coef
+      let bias_i = model.bias_i
+      let gamma_i = model.gamma_i
+      // radial basis function kernel
+      function rbf_kernel(support_vectors_i, input_vector, gamma_i) {
+          // Number of support vectors
+          const num_support_vectors = model.dim0
+          // Initialize the kernel values array
+          let kernel_values_i = new Array(num_support_vectors)
+          // Calculate the kernel values using nested for loops
+          for (let j = 0; j < num_support_vectors; j++) {
+              let sum_squares = 0
+              for (let i = 0; i < input_vector.length; i++) {
+                  //offset for reading 2D array as 1D vector
+                  let offset = j + (i*model.dim0)
+                  sum_squares += Math.pow(input_vector[i] - support_vectors_i[offset], 2)
+              }
+              kernel_values_i[j] = Math.exp(-gamma_i * sum_squares)
+          }
+          return kernel_values_i
+      }
+      let kernel_values_i = rbf_kernel(support_vectors_i, input_vector, gamma_i)
+      // Calculate the prediction using the regression function
+      let prediction = kernel_values_i.reduce((acc, kernel_value, index) => acc + coefficients_i[index] * kernel_value, 0)
+      prediction += bias_i
+      prediction_sum += prediction
+      // Feature weights and bias term (if needed for further use)
+      // let w = support_vectors_i.map((sv, index) => sv.map((value, j) => value * coefficients_i[index]))
+      // let b = bias_i
+    }
+    let prediction_mean = prediction_sum / models.length
+    return [acuteCoC , ROI_volML, prediction_mean]
   }
   openBtn.onclick = async function () {
     let input = document.createElement('input')
@@ -77,7 +114,9 @@ async function main() {
     input.click()
   }
   predictBtn.onclick = function () {
-    window.alert('Outcome prediction: 0.70289')
+    const [acuteCoC , ROI_volML, prediction] = neglect_predict()
+    const str = (`Given ${ROI_volML}ml lesion, and ${acuteCoC} acute CoC, predicted recovery is ${prediction}`)
+    window.alert(str)
   }
   aboutBtn.onclick = function () {
     window.alert('Drag and drop NIfTI images. Use pulldown menu to choose brainchop model')
@@ -88,7 +127,8 @@ async function main() {
   const defaults = {
     backColor: [0.4, 0.4, 0.4, 1],
     show3Dcrosshair: true,
-    onLocationChange: handleLocationChange
+    onLocationChange: handleLocationChange,
+    dragAndDropEnabled: false,
   }
   maskSlider.oninput = function () {
     nv1.setOpacity(1, this.value /255)
@@ -102,14 +142,13 @@ async function main() {
   const pca_mu = (await nv1.loadFromUrl('./pca_values_mu.nii.gz')).img
   async function loadRBFmodel() {
       const rbf = (await nv1.loadFromUrl('./models_5x10_diff.nii.gz')).img
-      let v = 0;
+      let v = 0
       const nModels = rbf[v++]
-      const models = []
+      const fmodels = []
       for (let i = 0; i < nModels; i++) {
         let model = {}
         model.dim0 = rbf[v++]
         model.dim1 = rbf[v++]
-        
         model.bias_i = rbf[v++]
         model.gamma_i = rbf[v++]
         model.SVs = new Float64Array(model.dim0 * model.dim1)
@@ -118,8 +157,9 @@ async function main() {
         model.sv_coef = new Float64Array(model.dim0)
         for (let j = 0; j < model.dim0; j++)
           model.sv_coef[j] = rbf[v++]
-        models.push(model)
+        fmodels.push(model)
       }
+      return fmodels
   }
   const models = await loadRBFmodel()
   nv1.opts.dragMode = nv1.dragModes.pan
@@ -134,7 +174,6 @@ async function main() {
   ])
   maskSlider.oninput()
   lesionSlider.oninput()
-  neglect_predict()
 }
 
 main()
diff --git a/niivue-neglect.png b/niivue-neglect.png
new file mode 100644
index 0000000..feac096
Binary files /dev/null and b/niivue-neglect.png differ