Add false risk detection

static/js/script.js

@@ -601,13 +601,20 @@ function showResults(results) {
 
 function generateResultsHTML(results) {
     const hasTumor = results.has_tumor;
+    const predictionStatus = results.prediction_status || (hasTumor ? 'TUMOR_DETECTED' : 'NO_TUMOR');
+    const isUncertainScan = predictionStatus === 'UNCERTAIN_SCAN';
+    const needsAttention = hasTumor || isUncertainScan;
     const confidence = (results.confidence * 100).toFixed(1);
     const report = results.detailed_report || {};
+    const safetyCheck = results.safety_check || {};
+    const uncertainty = safetyCheck.uncertainty || {};
 
     // Doctor's verdict based on results
-    const doctorVerdict = hasTumor 
+    const doctorVerdict = hasTumor
         ? `Based on my examination and the AI analysis, I've identified an abnormal tissue mass in the MRI scan. The ${confidence}% confidence level suggests this requires immediate attention. I recommend scheduling a follow-up consultation with a neurologist for further evaluation and to discuss treatment options.`
-        : `After carefully reviewing this MRI scan along with our AI analysis, I'm pleased to report that no tumor indicators were detected. The brain tissue appears healthy with normal density patterns. However, please continue with regular check-ups as recommended by your primary physician.`;
+        : isUncertainScan
+            ? `The primary model did not find a dominant tumor signal, but the secondary safety check flagged this scan as uncertain. Because low-confidence or high-variance negatives can hide subtle abnormalities, I recommend manual radiology review before treating this result as fully clear.`
+            : `After carefully reviewing this MRI scan along with our AI analysis, I'm pleased to report that no tumor indicators were detected. The brain tissue appears healthy with normal density patterns. However, please continue with regular check-ups as recommended by your primary physician.`;
 
     let html = `
         <!-- Share & Actions Bar -->
@@ -648,9 +655,9 @@ function generateResultsHTML(results) {
                         <span class="verdict-title">Senior Neuroradiologist</span>
                     </div>
                 </div>
-                <div class="verdict-stamp ${hasTumor ? 'concern' : 'clear'}">
-                    <i class="fas ${hasTumor ? 'fa-exclamation-circle' : 'fa-check-circle'}"></i>
-                    <span>${hasTumor ? 'Requires Attention' : 'All Clear'}</span>
+                <div class="verdict-stamp ${needsAttention ? 'concern' : 'clear'}">
+                    <i class="fas ${needsAttention ? 'fa-exclamation-circle' : 'fa-check-circle'}"></i>
+                    <span>${hasTumor ? 'Requires Attention' : (isUncertainScan ? 'Review Recommended' : 'All Clear')}</span>
                 </div>
             </div>
             <div class="verdict-content">
@@ -666,9 +673,9 @@ function generateResultsHTML(results) {
         </div>
 
         <div class="results-header">
-            <div class="result-badge ${hasTumor ? 'positive' : 'negative'}">
-                <i class="fas ${hasTumor ? 'fa-exclamation-triangle' : 'fa-check-circle'}"></i>
-                <span>${hasTumor ? 'Tumor Detected' : 'No Tumor Detected'}</span>
+            <div class="result-badge ${needsAttention ? 'positive' : 'negative'}">
+                <i class="fas ${needsAttention ? 'fa-exclamation-triangle' : 'fa-check-circle'}"></i>
+                <span>${hasTumor ? 'Tumor Detected' : (isUncertainScan ? 'Review Recommended' : 'No Tumor Detected')}</span>
             </div>
             <div class="confidence-score">${confidence}%</div>
             <div class="confidence-label">AI Confidence Score</div>
@@ -711,13 +718,13 @@ function generateResultsHTML(results) {
 
                 <!-- Report Summary Cards -->
                 <div class="report-summary-grid">
-                    <div class="summary-card ${hasTumor ? 'alert' : 'success'}">
+                    <div class="summary-card ${needsAttention ? 'alert' : 'success'}">
                         <div class="summary-icon">
-                            <i class="fas ${hasTumor ? 'fa-exclamation-triangle' : 'fa-check-circle'}"></i>
+                            <i class="fas ${needsAttention ? 'fa-exclamation-triangle' : 'fa-check-circle'}"></i>
                         </div>
                         <div class="summary-content">
                             <span class="summary-label">Detection Status</span>
-                            <span class="summary-value">${hasTumor ? 'Abnormality Detected' : 'No Abnormality'}</span>
+                            <span class="summary-value">${hasTumor ? 'Abnormality Detected' : (isUncertainScan ? 'Manual Review Recommended' : 'No Abnormality')}</span>
                         </div>
                     </div>
 
@@ -730,6 +737,18 @@ function generateResultsHTML(results) {
                             <span class="summary-value">${chars.confidence_score || confidence + '%'}</span>
                         </div>
                     </div>
+
+                    ${isUncertainScan ? `
+                    <div class="summary-card alert">
+                        <div class="summary-icon">
+                            <i class="fas fa-shield-alt"></i>
+                        </div>
+                        <div class="summary-content">
+                            <span class="summary-label">Safety Check</span>
+                            <span class="summary-value">Uncertain Scan</span>
+                        </div>
+                    </div>
+                    ` : ''}
 
                     ${hasTumor ? `
                     <div class="summary-card" style="border-left: 4px solid ${severity.color || '#f59e0b'}">
@@ -823,11 +842,11 @@ function generateResultsHTML(results) {
             // No tumor recommendations
             html += `
                 <div class="report-details-grid">
-                    <div class="details-card full-width success-card">
+                    <div class="details-card full-width ${isUncertainScan ? '' : 'success-card'}">
                         <h4><i class="fas fa-clipboard-list"></i> Recommendations</h4>
                         <ul class="recommendations-list">
                             ${(report.recommendations || []).map(rec => `
-                                <li><i class="fas fa-check"></i> ${rec}</li>
+                                <li><i class="fas ${isUncertainScan ? 'fa-chevron-right' : 'fa-check'}"></i> ${rec}</li>
                             `).join('')}
                         </ul>
                     </div>
@@ -853,6 +872,16 @@ function generateResultsHTML(results) {
                             <span class="meta-label">Ensemble</span>
                             <span class="meta-value">${metadata.ensemble_enabled ? 'Yes' : 'No'}</span>
                         </div>
+                        <div class="metadata-item">
+                            <span class="meta-label">Safety Status</span>
+                            <span class="meta-value">${hasTumor ? 'Not Needed' : (isUncertainScan ? 'Review Recommended' : 'Clear')}</span>
+                        </div>
+                        ${safetyCheck.applied && uncertainty.tumor_probability_variance !== undefined ? `
+                        <div class="metadata-item">
+                            <span class="meta-label">Uncertainty Variance</span>
+                            <span class="meta-value">${Number(uncertainty.tumor_probability_variance).toFixed(4)}</span>
+                        </div>
+                        ` : ''}
                         <div class="metadata-item">
                             <span class="meta-label">AI Version</span>
                             <span class="meta-value">${metadata.ai_version || '2.0.0'}</span>
@@ -1901,8 +1930,8 @@ async function loadScanHistory() {
                         }
                     </div>
                     <div class="history-item-info">
-                        <div class="history-item-result ${scan.has_tumor ? 'tumor-detected' : 'no-tumor'}">
-                            ${scan.has_tumor ? (scan.severity || 'Tumor Detected') : 'No Tumor'}
+                        <div class="history-item-result ${(scan.has_tumor || scan.prediction_status === 'UNCERTAIN_SCAN') ? 'tumor-detected' : 'no-tumor'}">
+                            ${scan.has_tumor ? (scan.severity || 'Tumor Detected') : (scan.prediction_status === 'UNCERTAIN_SCAN' ? 'Review Recommended' : 'No Tumor')}
                         </div>
                         <div class="history-item-confidence">
                             Confidence: ${(scan.confidence * 100).toFixed(1)}%

utilities.py

@@ -1,16 +1,46 @@
 
-import pandas as pd
 import numpy as np
-import seaborn as sns
 import cv2
 import tensorflow as tf
-import os 
-from skimage import io
+try:
+  from skimage import io
+except ImportError:
+  io = None
 from PIL import Image
 from tensorflow.keras import backend as K
 
 #creating a custom datagenerator:
 
+
+class MonteCarloDropout(tf.keras.layers.Dropout):
+  """Dropout layer that stays stochastic during inference for uncertainty sampling."""
+
+  def call(self, inputs, training = None):
+    return super().call(inputs, training = True)
+
+
+def model_supports_mc_dropout(model):
+  """Return True when the model contains dropout layers that can be sampled at inference time."""
+  return any(isinstance(layer, tf.keras.layers.Dropout) for layer in model.layers)
+
+
+def build_mc_dropout_model(model):
+  """
+  Clone a model so only dropout stays active during inference.
+  Batch normalization and the rest of the network remain in inference mode.
+  """
+  if not model_supports_mc_dropout(model):
+    return None
+
+  def clone_with_mc_dropout(layer):
+    if isinstance(layer, tf.keras.layers.Dropout):
+      return MonteCarloDropout.from_config(layer.get_config())
+    return layer.__class__.from_config(layer.get_config())
+
+  mc_model = tf.keras.models.clone_model(model, clone_function = clone_with_mc_dropout)
+  mc_model.set_weights(model.get_weights())
+  return mc_model
+
 class DataGenerator(tf.keras.utils.Sequence):
   def __init__(self, ids , mask, image_dir = './', batch_size = 16, img_h = 256, img_w = 256, shuffle = True):
 
@@ -59,6 +89,8 @@ def on_epoch_end(self):
 
   def __data_generation(self, list_ids, list_mask):
     'generate the data corresponding the indexes in a given batch of images'
+    if io is None:
+      raise ImportError('scikit-image is required for DataGenerator image loading.')
 
     # create empty arrays of shape (batch_size,height,width,depth) 
     #Depth is 3 for input and depth is taken as 1 for output becasue mask consist only of 1 channel.
@@ -109,6 +141,9 @@ def __data_generation(self, list_ids, list_mask):
 
 def prediction(test, model, model_seg):
   '''
+
+  if io is None:
+    raise ImportError('scikit-image is required for batch prediction image loading.')
   Predcition function which takes dataframe containing ImageID as Input and perform 2 type of prediction on the image
   Initially, image is passed through the classification network which predicts whether the image has defect or not, if the model
   is 99% sure that the image has no defect, then the image is labeled as no-defect, if the model is not sure, it passes the image to the
@@ -254,30 +289,30 @@ def precision_metric(y_true, y_pred):
 
 def predict_with_tta_classification(model, img):
     # Original
-    pred = model.predict(img)
+    pred = model.predict(img, verbose = 0)
 
     # Horizontal flip
     img_flip = np.flip(img, axis=2)
-    pred_flip = model.predict(img_flip)
+    pred_flip = model.predict(img_flip, verbose = 0)
 
     # Vertical flip
     img_vflip = np.flip(img, axis=1)
-    pred_vflip = model.predict(img_vflip)
+    pred_vflip = model.predict(img_vflip, verbose = 0)
 
     return (pred + pred_flip + pred_vflip) / 3
 
 def predict_with_tta_segmentation(model, img):
     # Original
-    pred = model.predict(img)
+    pred = model.predict(img, verbose = 0)
 
     # Horizontal flip
     img_flip = np.flip(img, axis=2)
-    pred_flip = model.predict(img_flip)
+    pred_flip = model.predict(img_flip, verbose = 0)
     pred_flip = np.flip(pred_flip, axis=2)
 
     # Vertical flip
     img_vflip = np.flip(img, axis=1)
-    pred_vflip = model.predict(img_vflip)
+    pred_vflip = model.predict(img_vflip, verbose = 0)
     pred_vflip = np.flip(pred_vflip, axis=1)
 
     return (pred + pred_flip + pred_vflip) / 3