Upload 3 files

app.py

@@ -0,0 +1,99 @@
+import streamlit as st
+import torch
+import torch.nn as nn
+import cv2
+import numpy as np
+from PIL import Image
+import albumentations as A
+from albumentations.pytorch import ToTensorV2
+import timm
+
+# --- 1. PAGE SETUP ---
+st.set_page_config(page_title="Medical AI Sonographer", page_icon="🩺", layout="centered")
+st.title("🩺 Dual-Head AI: Abdominal Ultrasound Diagnoser")
+st.write("Upload an abdominal ultrasound image. The AI will identify the organ and detect if it is normal or abnormal.")
+
+# --- 2. DEFINE THE AI ARCHITECTURE ---
+# (This must match exactly what you trained in Kaggle)
+class DualHeadConvNeXt(nn.Module):
+    def __init__(self, num_organs, num_statuses):
+        super().__init__()
+        # pretrained=False because we are loading our own weights!
+        self.backbone = timm.create_model('convnext_tiny', pretrained=False, num_classes=0)
+        in_features = self.backbone.num_features
+        self.organ_head = nn.Linear(in_features, num_organs)
+        self.status_head = nn.Linear(in_features, num_statuses)
+
+    def forward(self, x):
+        features = self.backbone(x)
+        return self.organ_head(features), self.status_head(features)
+
+# --- 3. LABELS & CACHING ---
+# Organ mapping based on your Kaggle Confusion Matrix
+ORGAN_LABELS = {
+    0: 'Spleen', 1: 'Pancreas', 2: 'Ovary', 3: 'Gallbladder', 
+    4: 'Portal Vein', 5: 'Liver', 6: 'Abdominal Aorta (AA)', 
+    7: 'Kidney', 8: 'UB / Prostate / Uterus', 9: 'Hepatic Vein', 10: 'Ascites'
+}
+STATUS_LABELS = {0: '✅ Normal (Healthy)', 1: '⚠️ Abnormal (Disease/Anomaly Detected)'}
+
+@st.cache_resource
+def load_model():
+    model = DualHeadConvNeXt(num_organs=11, num_statuses=2)
+    # map_location='cpu' is CRITICAL because web servers don't have Kaggle T4 GPUs
+    model.load_state_dict(torch.load('best_multitask_usg.pth', map_location=torch.device('cpu')))
+    model.eval()
+    return model
+
+model = load_model()
+
+# --- 4. IMAGE PREPROCESSING ---
+# Must match validation transforms from training
+transform = A.Compose([
+    A.Resize(384, 384),
+    A.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
+    ToTensorV2()
+])
+
+# --- 5. THE WEBSITE UI ---
+uploaded_file = st.file_uploader("Choose an Ultrasound Image (JPG/PNG)...", type=["jpg", "jpeg", "png"])
+
+if uploaded_file is not None:
+    # Show the image
+    image = Image.open(uploaded_file).convert('RGB')
+    st.image(image, caption='Uploaded Ultrasound', use_container_width=True)
+    
+    st.write("🧠 **AI is analyzing the image...**")
+    
+    # Preprocess
+    img_array = np.array(image)
+    tensor_img = transform(image=img_array)['image'].unsqueeze(0) # Add batch dimension
+    
+    # Predict
+    with torch.no_grad():
+        organ_pred, status_pred = model(tensor_img)
+        
+        organ_idx = organ_pred.argmax(dim=1).item()
+        status_idx = status_pred.argmax(dim=1).item()
+        
+        # Calculate confidences (probabilities)
+        organ_conf = torch.nn.functional.softmax(organ_pred, dim=1)[0][organ_idx].item() * 100
+        status_conf = torch.nn.functional.softmax(status_pred, dim=1)[0][status_idx].item() * 100
+
+    # --- 6. DISPLAY RESULTS BEAUTIFULLY ---
+    st.markdown("### 📊 AI Diagnosis Report")
+    
+    col1, col2 = st.columns(2)
+    with col1:
+        st.info(f"**Organ Detected:** \n### {ORGAN_LABELS[organ_idx]}")
+        st.write(f"Confidence: {organ_conf:.2f}%")
+        
+    with col2:
+        if status_idx == 0:
+            st.success(f"**Health Status:** \n### {STATUS_LABELS[status_idx]}")
+        else:
+            st.error(f"**Health Status:** \n### {STATUS_LABELS[status_idx]}")
+        st.write(f"Confidence: {status_conf:.2f}%")
+        
+    st.markdown("---")
+    st.caption("Disclaimer: This AI is for educational and research purposes only. It is not a substitute for professional medical diagnosis.")

best_multitask_usg (1).pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:247e02f1894742e5470d90b8e5b3734a37c13fa0cc213786800c4e0428eba17e
+size 111397305

requirements.txt

@@ -1,3 +1,8 @@
-altair
-pandas
-streamlit
+streamlit
+torch
+torchvision
+timm
+albumentations
+opencv-python-headless
+numpy
+Pillow