MoritzMMuller's picture
Update app.py
3ac60b6 verified
import os
import streamlit as st
from PIL import Image
import pandas as pd
from datetime import datetime
import requests
from geopy.geocoders import Nominatim
import folium
from streamlit_folium import st_folium
import cv2
import numpy as np
from huggingface_hub import snapshot_download
from transformers import (
AutoFeatureExtractor,
AutoModelForImageClassification,
ConvNextConfig,
pipeline,
)
st.set_page_config(page_title="Skin Cancer Dashboard", layout="wide")
# --- Configuration ---
# Ensure you have set your Hugging Face token as an environment variable:
#export HF_TOKEN="YOUR_TOKEN_HERE"
MODEL_NAME = "Anwarkh1/Skin_Cancer-Image_Classification"
LLM_NAME = "google/flan-t5-xl"
HF_TOKEN = os.environ.get("HF_TOKEN")
DATA_DIR = "data/harvard_dataset" # Path where you download and unpack the Harvard Dataverse dataset
DIARY_CSV = "diary.csv"
# Initialize session state defaults
if 'initialized' not in st.session_state:
st.session_state['label'] = None
st.session_state['score'] = None
st.session_state['mole_id'] = ''
st.session_state['geo_location'] = ''
st.session_state['chat_history'] = []
st.session_state['initialized'] = True
# Initialize geolocator for free geocoding
geolocator = Nominatim(user_agent="skin-dashboard", timeout = 10)
@st.cache_resource
def load_image_model(token: str):
return pipeline(
"image-classification",
feature_extractor=AutoFeatureExtractor.from_pretrained(
MODEL_NAME,
#subfolder="Skin_Cancer-Image_Classification",
use_auth_token=token
),
model=AutoModelForImageClassification.from_pretrained(
MODEL_NAME,
#subfolder="Skin_Cancer-Image_Classification",
use_auth_token=token
),
device=0 # or -1 for CPU
)
@st.cache_resource
def load_llm(token: str):
return pipeline(
"text2text-generation",
model=LLM_NAME,
device_map="auto", # or device=0 for single GPU / -1 for CPU
max_length=10000,
num_beams=5,
no_repeat_ngram_size=2,
early_stopping=True,
)
classifier = load_image_model(HF_TOKEN) if HF_TOKEN else None
explainer = load_llm(HF_TOKEN) if HF_TOKEN else None
# --- Diary Init ----
if not os.path.exists(DIARY_CSV):
pd.DataFrame(
columns=["timestamp", "image_path", "mole_id", "geo_location", "label", "score",
"body_location", "prior_consultation", "pain", "itch"]
).to_csv(DIARY_CSV, index=False)
# --- Save entry helper
def save_entry(img_path: str, mole_id: str, geo_location: str,
label: str, score: float,
body_location: str, prior_consult: str, pain: str, itch: str):
df = pd.read_csv(DIARY_CSV)
entry = {
"timestamp": datetime.now().isoformat(),
"image_path": img_path,
"mole_id": mole_id,
"geo_location": geo_location,
"label": label,
"score": float(score),
"body_location": body_location,
"prior_consultation": prior_consult,
"pain": pain,
"itch": itch
}
df.loc[len(df)] = entry
df.to_csv(DIARY_CSV, index=False)
# --- Preprocessing Functions ---
def remove_hair(img: np.ndarray) -> np.ndarray:
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (17, 17))
blackhat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT, kernel)
_, mask = cv2.threshold(blackhat, 10, 255, cv2.THRESH_BINARY)
return cv2.inpaint(img, mask, 1, cv2.INPAINT_TELEA)
def preprocess(img: Image.Image, size: int = 224) -> Image.Image:
arr = np.array(img)
bgr = cv2.cvtColor(arr, cv2.COLOR_RGB2BGR)
bgr = remove_hair(bgr)
bgr = cv2.bilateralFilter(bgr, d=9, sigmaColor=75, sigmaSpace=75)
lab = cv2.cvtColor(bgr, cv2.COLOR_BGR2LAB)
l, a, b = cv2.split(lab)
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
cl = clahe.apply(l)
merged = cv2.merge((cl, a, b))
bgr = cv2.cvtColor(merged, cv2.COLOR_LAB2BGR)
h, w = bgr.shape[:2]
scale = size / max(h, w)
nh, nw = int(h*scale), int(w*scale)
resized = cv2.resize(bgr, (nw, nh), interpolation=cv2.INTER_AREA)
canvas = np.full((size, size, 3), 128, dtype=np.uint8)
top, left = (size-nh)//2, (size-nw)//2
canvas[top:top+nh, left:left+nw] = resized
rgb = cv2.cvtColor(canvas, cv2.COLOR_BGR2RGB)
return Image.fromarray(rgb)
# -----Streamlit layout ----
st.title("🩺 Skin Cancer Recognition Dashboard")
menu = ["Scan Mole","Chat","Diary", "Dataset Explorer"]
choice = st.sidebar.selectbox("Navigation", menu)
# --- Initialize Scan a Mole ---
if choice == "Scan Mole":
st.header("🔍 Scan a Mole")
if not classifier:
st.error("Missing HF_TOKEN.")
st.stop()
upload = st.file_uploader("Upload a skin image", type=["jpg","jpeg","png"])
if not upload:
st.info("Please upload an image to begin.")
st.stop()
raw = Image.open(upload).convert("RGB")
st.image(raw, caption="Original", use_container_width=True)
proc = preprocess(raw)
st.image(proc, caption="Preprocessed", use_container_width=True)
mole = st.text_input("Mole ID")
city = st.text_input("Geographic location")
body = st.selectbox("Body location", ["Face","Scalp","Neck","Chest","Back","Arm","Hand","Leg","Foot","Other"])
prior = st.radio("Prior consult?", ["Yes","No"], horizontal=True)
pain = st.radio("Pain?", ["Yes","No"], horizontal=True)
itch = st.radio("Itch?", ["Yes","No"], horizontal=True)
if st.button("Classify"):
if not mole or not city:
st.error("Enter ID and location.")
else:
with st.spinner("Analyzing..."):
out = classifier(proc)
lbl, scr = out[0]["label"], out[0]["score"]
save_dir = os.path.join("scans", f"{mole}_{datetime.now().timestamp()}.png")
os.makedirs(os.path.dirname(save_dir), exist_ok=True)
raw.save(save_dir)
save_entry(save_dir, mole, city, lbl, scr, body, prior, pain, itch)
st.session_state.update({
'label': lbl,
'score': scr,
'mole_id': mole,
'geo_location': city
})
if st.session_state['label']:
st.success(f"Prediction: {st.session_state['label']} (score {st.session_state['score']:.2f})")
if explainer:
with st.spinner("Explaining..."):
text = explainer(f"Explain {st.session_state['label']} and recommendation.")[0]['generated_text']
st.markdown("### Explanation"); st.write(text)
loc = geolocator.geocode(st.session_state['geo_location'])
if loc:
m = folium.Map([loc.latitude, loc.longitude], zoom_start=12)
folium.Marker([loc.latitude, loc.longitude], "You").add_to(m)
resp = requests.post(
"https://overpass-api.de/api/interpreter",
data={"data": f"[out:json];node(around:5000,{loc.latitude},{loc.longitude})[~\"^(amenity|healthcare)$\"~\"clinic|doctors\"];out;"}
)
for el in resp.json().get('elements', []):
tags = el.get('tags', {});
lat = el.get('lat') or el['center']['lat']; lon = el.get('lon') or el['center']['lon']
folium.Marker([lat, lon], tags.get('name','Clinic')).add_to(m)
st.markdown("### Nearby Clinics"); st_folium(m, width=700)
# --- Chat Tab ---
elif choice == "Chat":
st.header("💬 Follow-Up Chat")
if not st.session_state['label']:
st.info("Please perform a scan first in the 'Scan Mole' tab.")
else:
lbl = st.session_state['label']
scr = st.session_state['score']
mid = st.session_state['mole_id']
gloc = st.session_state['geo_location']
st.markdown(f"**Context:** prediction for **{mid}** at **{gloc}** is **{lbl}** (confidence {scr:.2f}).")
# New user message comes first for immediate loop
user_q = st.chat_input("Ask a follow-up question:", key="chat_input")
if user_q and explainer:
st.session_state['chat_history'].append({'role':'user','content':user_q})
system_p = "You are a dermatology assistant. Provide concise medical advice without clarifying questions."
tpl = (
f"{system_p}\nContext: prediction is {lbl} with confidence {scr:.2f}.\n"
f"User: {user_q}\nAssistant:"
)
with st.spinner("Generating response..."):
reply = explainer(tpl)[0]['generated_text']
st.session_state['chat_history'].append({'role':'assistant','content':reply})
# Display the updated chat history
for msg in st.session_state['chat_history']:
prefix = 'You' if msg['role']=='user' else 'AI'
st.markdown(f"**{prefix}:** {msg['content']}")
# --- Diary Page ---
elif choice == "Diary":
st.header("📖 Skin Cancer Diary")
df = pd.read_csv(DIARY_CSV)
df['timestamp'] = pd.to_datetime(df['timestamp'])
if df.empty:
st.info("No diary entries yet.")
else:
mole_ids = sorted(df['mole_id'].unique())
sel = st.selectbox("Select Mole to View", ['All'] + mole_ids, key="diary_sel")
if sel == 'All':
# Display moles in columns (max 3 per row)
chunks = [mole_ids[i:i+3] for i in range(0, len(mole_ids), 3)]
for group in chunks:
cols = st.columns(len(group))
for col, mid in zip(cols, group):
with col:
st.subheader(mid)
entries = df[df['mole_id'] == mid].sort_values('timestamp')
# Show image timeline
for _, row in entries.iterrows():
if os.path.exists(row['image_path']):
st.image(
row['image_path'],
width=150,
caption=f"{row['timestamp'].strftime('%Y-%m-%d')}{row['score']:.2f}"
)
st.write(f"Total scans: {len(entries)}")
else:
# Detailed view for a single mole
entries = df[df['mole_id'] == sel].sort_values('timestamp')
if entries.empty:
st.warning(f"No entries for {sel}.")
else:
# Score over time
st.line_chart(entries.set_index('timestamp')['score'])
st.markdown("#### Image Timeline")
for _, row in entries.iterrows():
if os.path.exists(row['image_path']):
st.image(
row['image_path'],
width=200,
caption=(
f"{row['timestamp'].strftime('%Y-%m-%d %H:%M')} — "
f"Score: {row['score']:.2f}"
)
)
st.markdown("#### Details")
st.dataframe(
entries[
['timestamp','geo_location','label','score',
'body_location','prior_consultation','pain','itch']
]
.rename(columns={
'timestamp':'Time','geo_location':'Location',
'label':'Diagnosis','score':'Confidence',
'body_location':'Body Part','prior_consultation':'Prior Consult',
'pain':'Pain','itch':'Itch'
})
.sort_values('Time', ascending=False)
)
else:
st.header("📂 Dataset Explorer")
st.write("Preview images from the Harvard Skin Cancer Dataset")
# pick up to 15 image files
image_files = [
f for f in os.listdir(DATA_DIR)
if os.path.isfile(os.path.join(DATA_DIR, f))
and f.lower().endswith((".jpg", ".jpeg", ".png"))
][:15]
for i in range(0, len(image_files), 3):
cols = st.columns(3)
for col, fn in zip(cols, image_files[i : i + 3]):
path = os.path.join(DATA_DIR, fn)
img = Image.open(path)
col.image(img, use_container_width=True)
col.caption(fn)
st.sidebar.markdown("---")
st.sidebar.write("Dataset powered by Harvard Dataverse [DBW86T]")
st.sidebar.write(f"Model: {MODEL_NAME}")
st.sidebar.write(f"LLM: {LLM_NAME}")
if __name__ == '__main__':
st.write()