File size: 12,834 Bytes
cf9cd6c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2a50d56
4bae555
 
cf9cd6c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2a50d56
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
"""
CliniScan Local UI β€” Backend
Run:  pip install fastapi uvicorn python-multipart torch torchvision albumentations opencv-python-headless ultralytics grad-cam reportlab
Then: uvicorn app:app --reload --port 8000
"""

import os, io, time, warnings, base64, json
warnings.filterwarnings("ignore")

import numpy as np
import cv2
import torch
import torch.nn as nn
from torchvision.models import efficientnet_b0
import albumentations as A
from albumentations.pytorch import ToTensorV2

from fastapi import FastAPI, File, UploadFile, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse, HTMLResponse
from fastapi.staticfiles import StaticFiles
from typing import List, Optional
import tempfile

BASE = os.path.dirname(os.path.abspath(__file__))
CLS_MODEL_PATH  = os.path.join(BASE, "Exp2_AdamW.pt")
YOLO_MODEL_PATH = os.path.join(BASE, "last.pt")
CLASS_NAMES = [
    "Aortic enlargement","Atelectasis","Calcification","Cardiomegaly",
    "Consolidation","ILD","Infiltration","Lung Opacity",
    "Nodule/Mass","Other lesion","Pleural effusion",
    "Pleural thickening","Pneumothorax","Pulmonary fibrosis",
]
NUM_CLASSES = len(CLASS_NAMES)
IMG_SIZE    = 224
DEVICE      = torch.device("cuda" if torch.cuda.is_available() else "cpu")

eval_tfm = A.Compose([
    A.Resize(IMG_SIZE, IMG_SIZE),
    A.Normalize(mean=[0.485,0.456,0.406], std=[0.229,0.224,0.225]),
    ToTensorV2(),
])

# ── Load models ──────────────────────────────────────────────
cls_model  = None
yolo_model = None

def load_models():
    global cls_model, yolo_model
    try:
        m = efficientnet_b0(weights=None)
        m.classifier[1] = nn.Linear(m.classifier[1].in_features, NUM_CLASSES)
        if os.path.exists(CLS_MODEL_PATH):
            ckpt = torch.load(CLS_MODEL_PATH, map_location=DEVICE, weights_only=False)
            state = ckpt['model'] if 'model' in ckpt else ckpt
            m.load_state_dict(state)
            print(f"βœ… Classification model loaded")
        else:
            print(f"⚠️  Model not found: {CLS_MODEL_PATH}")
            print("   Running in demo mode with random weights")
        cls_model = m.to(DEVICE)
        cls_model.eval()
    except Exception as e:
        print(f"❌ Cls error: {e}")
    try:
        from ultralytics import YOLO
        if os.path.exists(YOLO_MODEL_PATH):
            yolo_model = YOLO(YOLO_MODEL_PATH)
            print(f"βœ… YOLO detection model loaded")
        else:
            print(f"⚠️  YOLO not found: {YOLO_MODEL_PATH}")
    except Exception as e:
        print(f"❌ YOLO error: {e}")

load_models()

# ── Helpers ──────────────────────────────────────────────────
def preprocess(file_bytes):
    arr = np.frombuffer(file_bytes, np.uint8)
    img = cv2.imdecode(arr, cv2.IMREAD_GRAYSCALE)
    if img is None:
        raise HTTPException(400, "Invalid image")
    h, w = img.shape[:2]
    img  = cv2.resize(img[int(h*.02):int(h*.98), int(w*.02):int(w*.98)], (w, h))
    return cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

def classify(rgb):
    if cls_model is None:
        import random; return [(c, round(random.uniform(.05,.90),4)) for c in CLASS_NAMES]
    t = eval_tfm(image=rgb)["image"].unsqueeze(0).to(DEVICE)
    with torch.no_grad():
        p = torch.sigmoid(cls_model(t)).cpu().numpy()[0]
    return [(CLASS_NAMES[i], float(round(p[i],4))) for i in range(NUM_CLASSES)]

def detect(rgb):
    if yolo_model is None: return [], None
    try:
        tmp = os.path.join(tempfile.gettempdir(), "_cs.png")
        cv2.imwrite(tmp, cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR))
        res   = yolo_model.predict(tmp, conf=0.25, verbose=False)[0]
        boxes = []
        if res.boxes:
            for b in res.boxes:
                x1,y1,x2,y2 = b.xyxy[0].tolist(); cls = int(b.cls[0])
                boxes.append({"x1":round(x1),"y1":round(y1),"x2":round(x2),"y2":round(y2),
                               "conf":round(float(b.conf[0]),3),
                               "class":CLASS_NAMES[cls] if cls<NUM_CLASSES else f"cls{cls}"})
        vis = res.plot()
        _, buf = cv2.imencode(".png", vis)
        return boxes, base64.b64encode(buf).decode()
    except Exception as e:
        print(f"Det error: {e}"); return [], None

def gradcam(rgb, target=None):
    if cls_model is None: return None
    try:
        from pytorch_grad_cam import GradCAM
        from pytorch_grad_cam.utils.image import show_cam_on_image
        from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
        cam  = GradCAM(model=cls_model, target_layers=[cls_model.features[-1]])
        t    = eval_tfm(image=rgb)["image"].unsqueeze(0).to(DEVICE)
        with torch.no_grad():
            p = torch.sigmoid(cls_model(t)).cpu().numpy()[0]
        top  = int(np.argmax(p)) if target is None else target
        gc   = cam(t, [ClassifierOutputTarget(top)])[0]
        r224 = cv2.resize(rgb, (IMG_SIZE, IMG_SIZE))
        ov   = show_cam_on_image(r224.astype(np.float32)/255.0, gc, use_rgb=True)
        _, buf = cv2.imencode(".png", cv2.cvtColor(ov, cv2.COLOR_RGB2BGR))
        return base64.b64encode(buf).decode()
    except Exception as e:
        print(f"CAM error: {e}"); return None

def make_pdf(preds, fname, boxes):
    from reportlab.lib.pagesizes import letter
    from reportlab.lib import colors
    from reportlab.lib.styles import getSampleStyleSheet
    from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Table, TableStyle
    from reportlab.lib.units import inch
    buf = io.BytesIO()
    doc = SimpleDocTemplate(buf, pagesize=letter,
                             rightMargin=inch, leftMargin=inch,
                             topMargin=inch, bottomMargin=inch)
    st = getSampleStyleSheet(); story = []
    story.append(Paragraph("CliniScan β€” AI Chest X-Ray Report", st["Title"]))
    story.append(Spacer(1,.15*inch))
    story.append(Paragraph(f"File: {fname}  |  {time.strftime('%Y-%m-%d %H:%M')}", st["Normal"]))
    story.append(Paragraph("Research use only β€” not a clinical diagnosis.", st["Normal"]))
    story.append(Spacer(1,.25*inch))
    story.append(Paragraph("Classification Results", st["Heading2"]))
    data = [["Disease","Confidence","Status"]]
    for n,c in sorted(preds, key=lambda x:-x[1]):
        data.append([n, f"{c:.1%}", "DETECTED" if c>=.5 else "Normal"])
    t = Table(data, colWidths=[3*inch,1.5*inch,1.5*inch])
    t.setStyle(TableStyle([
        ("BACKGROUND",(0,0),(-1,0),colors.HexColor("#1F4E79")),
        ("TEXTCOLOR",(0,0),(-1,0),colors.white),
        ("FONTNAME",(0,0),(-1,0),"Helvetica-Bold"),
        ("FONTSIZE",(0,0),(-1,-1),10),
        ("ROWBACKGROUNDS",(0,1),(-1,-1),[colors.white,colors.HexColor("#F2F2F2")]),
        ("GRID",(0,0),(-1,-1),.5,colors.HexColor("#CCCCCC")),
        ("TOPPADDING",(0,0),(-1,-1),5),("BOTTOMPADDING",(0,0),(-1,-1),5),
    ]))
    story.append(t)
    if boxes:
        story.append(Spacer(1,.2*inch))
        story.append(Paragraph("Detection Bounding Boxes", st["Heading2"]))
        dd = [["Finding","Confidence","Location"]]
        for b in boxes:
            dd.append([b["class"],f"{b['conf']:.1%}",
                       f"({b['x1']},{b['y1']}) to ({b['x2']},{b['y2']})"])
        dt = Table(dd, colWidths=[2.5*inch,1.5*inch,2*inch])
        dt.setStyle(TableStyle([
            ("BACKGROUND",(0,0),(-1,0),colors.HexColor("#2E75B6")),
            ("TEXTCOLOR",(0,0),(-1,0),colors.white),
            ("FONTSIZE",(0,0),(-1,-1),10),
            ("GRID",(0,0),(-1,-1),.5,colors.HexColor("#CCCCCC")),
            ("ROWBACKGROUNDS",(0,1),(-1,-1),[colors.white,colors.HexColor("#F2F2F2")]),
            ("TOPPADDING",(0,0),(-1,-1),5),("BOTTOMPADDING",(0,0),(-1,-1),5),
        ]))
        story.append(dt)
    doc.build(story); buf.seek(0); return buf

# ── FastAPI app ──────────────────────────────────────────────
app = FastAPI(title="CliniScan")
app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])

# Serve the HTML frontend
@app.get("/", response_class=HTMLResponse)
async def serve_ui():
    html_path = os.path.join(os.path.dirname(__file__), "index.html")
    if os.path.exists(html_path):
        with open(html_path, "r", encoding="utf-8") as f:
            return f.read()
    return "<h1>index.html not found β€” place it in the same folder as app.py</h1>"

@app.get("/health")
def health():
    return {"status":"ok","cls_model":cls_model is not None,
            "yolo_model":yolo_model is not None,"device":str(DEVICE)}

@app.get("/model-info")
def model_info():
    return {
        "classification":{"name":"EfficientNet-B0 (Exp2_AdamW)","backbone":"EfficientNet-B0",
                          "optimizer":"AdamW","lr":"5e-4","epochs":12,
                          "val_auc":0.9213,"baseline_auc":0.8858,"improvement":"+3.55%"},
        "detection":{"name":"YOLOv8s (D2_LowLR)","model":"YOLOv8s",
                     "optimizer":"AdamW","lr":"5e-4","epochs":15,
                     "map50":0.0592,"map50_95":0.0251,"baseline_map50":0.0658},
        "cls_experiments":[
            {"name":"Exp1_Adam",    "auc":0.9104,"optimizer":"Adam",  "lr":"1e-3"},
            {"name":"Exp2_AdamW",   "auc":0.9213,"optimizer":"AdamW", "lr":"5e-4"},
            {"name":"Exp3_SGD",     "auc":0.9094,"optimizer":"SGD",   "lr":"1e-2"},
            {"name":"Exp4_DenseNet","auc":0.8304,"optimizer":"Adam",  "lr":"1e-3"},
            {"name":"Baseline M2",  "auc":0.8858,"optimizer":"Adam",  "lr":"1e-3"},
        ],
        "det_experiments":[
            {"name":"D1_YOLOv8m",  "map50":0.0463,"prec":0.6951,"rec":0.0527},
            {"name":"D2_LowLR",    "map50":0.0592,"prec":0.3683,"rec":0.0725},
            {"name":"D3_LowThresh","map50":0.0574,"prec":0.1749,"rec":0.0624},
            {"name":"D4_SGD",      "map50":0.0446,"prec":0.5503,"rec":0.0436},
            {"name":"Baseline M2", "map50":0.0658,"prec":0.3793,"rec":0.0821},
        ],
    }

@app.post("/predict")
async def predict(
    file: UploadFile = File(...),
    threshold: float = 0.5,
    run_detection: bool = True,
    run_gradcam: bool = True,
):
    t0    = time.time()
    data  = await file.read()
    rgb   = preprocess(data)
    preds = classify(rgb)

    boxes, det_vis = [], None
    if run_detection: boxes, det_vis = detect(rgb)

    cam_b64 = None
    if run_gradcam: cam_b64 = gradcam(rgb)

    _, buf   = cv2.imencode(".png", cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR))
    orig_b64 = base64.b64encode(buf).decode()
    detected = sorted([{"disease":n,"confidence":c} for n,c in preds if c>=threshold],
                      key=lambda x:-x["confidence"])
    return {
        "filename":    file.filename,
        "predictions": [{"disease":n,"confidence":c} for n,c in preds],
        "detected":    detected,
        "boxes":       boxes,
        "images":{"original":orig_b64,"detection":det_vis,"gradcam":cam_b64},
        "threshold":   threshold,
        "time_ms":     round((time.time()-t0)*1000),
    }

@app.post("/batch")
async def batch_predict(files: List[UploadFile] = File(...), threshold: float = 0.5):
    results = []
    for f in files:
        try:
            data  = await f.read()
            rgb   = preprocess(data)
            preds = classify(rgb)
            det   = [n for n,c in preds if c>=threshold]
            results.append({"filename":f.filename,"status":"success",
                             "detected":det,"findings":len(det),
                             "top_conf":round(max(c for _,c in preds),4),
                             "predictions":[{"disease":n,"confidence":c} for n,c in preds]})
        except Exception as e:
            results.append({"filename":f.filename,"status":"error","error":str(e)})
    return {"total":len(results),
            "processed":sum(1 for r in results if r["status"]=="success"),
            "errors":sum(1 for r in results if r["status"]=="error"),
            "results":results}

@app.post("/report")
async def generate_report(file: UploadFile = File(...)):
    data  = await file.read()
    rgb   = preprocess(data)
    preds = classify(rgb)
    boxes, _ = detect(rgb)
    pdf   = make_pdf(preds, file.filename, boxes)
    return StreamingResponse(pdf, media_type="application/pdf",
        headers={"Content-Disposition":"attachment; filename=cliniscan_report.pdf"})

if __name__ == "__main__":
    import uvicorn
    port = int(os.environ.get("PORT", 7860))
    host = "0.0.0.0" if os.environ.get("PORT") else "127.0.0.1"
    uvicorn.run("app:app", host=host, port=port, reload=False)