Spaces:
Sleeping
Sleeping
Upload 4 files
Browse files- ada_learn_malben.pkl +3 -0
- ada_learn_skin_norm2000.pkl +3 -0
- app.py +120 -0
- requirements.txt +7 -0
ada_learn_malben.pkl
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:441760abd9a3a6b143917c68cf1709a1d96323fa334b4952009f50a34ddb57c2
|
3 |
+
size 87548073
|
ada_learn_skin_norm2000.pkl
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:820252747a54e3e7494fdc86c998d1d67996cb31bde244ddd4260307e80dd819
|
3 |
+
size 87739881
|
app.py
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import ViTImageProcessor, ViTForImageClassification, pipeline
|
3 |
+
from fastai.learner import load_learner
|
4 |
+
from fastai.vision.core import PILImage
|
5 |
+
from PIL import Image
|
6 |
+
import matplotlib.pyplot as plt
|
7 |
+
import numpy as np
|
8 |
+
import gradio as gr
|
9 |
+
import io
|
10 |
+
import base64
|
11 |
+
|
12 |
+
# 🔹 Modelo ViT desde Hugging Face (HAM10000)
|
13 |
+
MODEL_NAME = "ahishamm/vit-base-HAM-10000-sharpened-patch-32"
|
14 |
+
feature_extractor = ViTImageProcessor.from_pretrained(MODEL_NAME)
|
15 |
+
model_vit = ViTForImageClassification.from_pretrained(MODEL_NAME)
|
16 |
+
model_vit.eval()
|
17 |
+
|
18 |
+
# 🔹 Modelos Fast.ai desde archivo local
|
19 |
+
model_malignancy = load_learner("ada_learn_malben.pkl")
|
20 |
+
model_norm2000 = load_learner("ada_learn_skin_norm2000.pkl")
|
21 |
+
|
22 |
+
# 🔹 Modelo binario ISIC preentrenado (alta fiabilidad)
|
23 |
+
classifier_isic = pipeline("image-classification", model="VRJBro/skin-cancer-detection")
|
24 |
+
|
25 |
+
# 🔹 Clases y niveles de riesgo
|
26 |
+
CLASSES = [
|
27 |
+
"Queratosis actínica / Bowen", "Carcinoma células basales",
|
28 |
+
"Lesión queratósica benigna", "Dermatofibroma",
|
29 |
+
"Melanoma maligno", "Nevus melanocítico", "Lesión vascular"
|
30 |
+
]
|
31 |
+
RISK_LEVELS = {
|
32 |
+
0: {'level': 'Moderado', 'color': '#ffaa00', 'weight': 0.6},
|
33 |
+
1: {'level': 'Alto', 'color': '#ff4444', 'weight': 0.8},
|
34 |
+
2: {'level': 'Bajo', 'color': '#44ff44', 'weight': 0.1},
|
35 |
+
3: {'level': 'Bajo', 'color': '#44ff44', 'weight': 0.1},
|
36 |
+
4: {'level': 'Crítico', 'color': '#cc0000', 'weight': 1.0},
|
37 |
+
5: {'level': 'Bajo', 'color': '#44ff44', 'weight': 0.1},
|
38 |
+
6: {'level': 'Bajo', 'color': '#44ff44', 'weight': 0.1}
|
39 |
+
}
|
40 |
+
|
41 |
+
def analizar_lesion_combined(img):
|
42 |
+
img_fastai = PILImage.create(img)
|
43 |
+
|
44 |
+
# 🔹 ViT prediction
|
45 |
+
inputs = feature_extractor(img, return_tensors="pt")
|
46 |
+
with torch.no_grad():
|
47 |
+
outputs = model_vit(**inputs)
|
48 |
+
probs_vit = outputs.logits.softmax(dim=-1).cpu().numpy()[0]
|
49 |
+
pred_idx_vit = int(np.argmax(probs_vit))
|
50 |
+
pred_class_vit = CLASSES[pred_idx_vit]
|
51 |
+
confidence_vit = probs_vit[pred_idx_vit]
|
52 |
+
|
53 |
+
# 🔹 Fast.ai predictions
|
54 |
+
pred_fast_malignant, _, probs_fast_mal = model_malignancy.predict(img_fastai)
|
55 |
+
prob_malignant = float(probs_fast_mal[1])
|
56 |
+
pred_fast_type, _, probs_fast_type = model_norm2000.predict(img_fastai)
|
57 |
+
|
58 |
+
# 🔹 ISIC binary classification (modelo 4)
|
59 |
+
result_isic = classifier_isic(img)
|
60 |
+
pred_isic = result_isic[0]['label']
|
61 |
+
confidence_isic = result_isic[0]['score']
|
62 |
+
|
63 |
+
# 🔹 Gráfico ViT
|
64 |
+
colors_bars = [RISK_LEVELS[i]['color'] for i in range(7)]
|
65 |
+
fig, ax = plt.subplots(figsize=(8, 3))
|
66 |
+
ax.bar(CLASSES, probs_vit*100, color=colors_bars)
|
67 |
+
ax.set_title("Probabilidad ViT por tipo de lesión")
|
68 |
+
ax.set_ylabel("Probabilidad (%)")
|
69 |
+
ax.set_xticks(np.arange(len(CLASSES)))
|
70 |
+
ax.set_xticklabels(CLASSES, rotation=45, ha='right')
|
71 |
+
ax.grid(axis='y', alpha=0.2)
|
72 |
+
plt.tight_layout()
|
73 |
+
buf = io.BytesIO()
|
74 |
+
plt.savefig(buf, format="png")
|
75 |
+
plt.close(fig)
|
76 |
+
img_bytes = buf.getvalue()
|
77 |
+
img_b64 = base64.b64encode(img_bytes).decode("utf-8")
|
78 |
+
html_chart = f'<img src="data:image/png;base64,{img_b64}" style="max-width:100%"/>'
|
79 |
+
|
80 |
+
# 🔹 Informe HTML
|
81 |
+
informe = f"""
|
82 |
+
<div style="font-family:sans-serif; max-width:800px; margin:auto">
|
83 |
+
<h2>🧪 Diagnóstico por 4 modelos de IA</h2>
|
84 |
+
<table style="border-collapse: collapse; width:100%; font-size:16px">
|
85 |
+
<tr><th style="text-align:left">🔍 Modelo</th><th>Resultado</th><th>Confianza</th></tr>
|
86 |
+
<tr><td>🧠 ViT (transformer)</td><td><b>{pred_class_vit}</b></td><td>{confidence_vit:.1%}</td></tr>
|
87 |
+
<tr><td>🧬 Fast.ai (clasificación)</td><td><b>{pred_fast_type}</b></td><td>N/A</td></tr>
|
88 |
+
<tr><td>⚠️ Fast.ai (malignidad)</td><td><b>{"Maligno" if prob_malignant > 0.5 else "Benigno"}</b></td><td>{prob_malignant:.1%}</td></tr>
|
89 |
+
<tr><td>🔬 ISIC binario</td><td><b>{pred_isic.capitalize()}</b></td><td>{confidence_isic:.1%}</td></tr>
|
90 |
+
</table>
|
91 |
+
<br>
|
92 |
+
<b>🩺 Recomendación automática:</b><br>
|
93 |
+
"""
|
94 |
+
|
95 |
+
cancer_risk_score = sum(probs_vit[i] * RISK_LEVELS[i]['weight'] for i in range(7))
|
96 |
+
if prob_malignant > 0.7 or cancer_risk_score > 0.6 or (pred_isic == "cancerous" and confidence_isic > 0.9):
|
97 |
+
informe += "🚨 <b>CRÍTICO</b> – Derivación urgente a oncología dermatológica"
|
98 |
+
elif prob_malignant > 0.4 or cancer_risk_score > 0.4 or (pred_isic == "cancerous"):
|
99 |
+
informe += "⚠️ <b>ALTO RIESGO</b> – Consulta con dermatólogo en 7 días"
|
100 |
+
elif cancer_risk_score > 0.2:
|
101 |
+
informe += "📋 <b>RIESGO MODERADO</b> – Evaluación programada (2-4 semanas)"
|
102 |
+
else:
|
103 |
+
informe += "✅ <b>BAJO RIESGO</b> – Seguimiento de rutina (3-6 meses)"
|
104 |
+
|
105 |
+
informe += "</div>"
|
106 |
+
|
107 |
+
return informe, html_chart
|
108 |
+
|
109 |
+
# 🔹 Interfaz Gradio actualizada
|
110 |
+
demo = gr.Interface(
|
111 |
+
fn=analizar_lesion_combined,
|
112 |
+
inputs=gr.Image(type="pil", label="Sube una imagen de la lesión"),
|
113 |
+
outputs=[gr.HTML(label="Informe combinado"), gr.HTML(label="Gráfico ViT")],
|
114 |
+
title="Detector de Lesiones Cutáneas (ViT + Fast.ai + ISIC)",
|
115 |
+
description="Comparación entre ViT transformer (HAM10000), dos modelos Fast.ai y un clasificador binario ISIC con alta precisión.",
|
116 |
+
flagging_mode="never"
|
117 |
+
)
|
118 |
+
|
119 |
+
if __name__ == "__main__":
|
120 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch==2.0.1
|
2 |
+
torchvision==0.15.2
|
3 |
+
fastai==2.7.12
|
4 |
+
transformers==4.41.2
|
5 |
+
numpy<2
|
6 |
+
gradio==5.38.1
|
7 |
+
matplotlib
|