Imene-ee commited on
Commit
a1f3d11
·
verified ·
1 Parent(s): 3b67e5e

Upload 4 files

Browse files
Files changed (4) hide show
  1. LICENSE +21 -0
  2. app.py +195 -0
  3. packages.txt +6 -0
  4. requirements.txt +11 -0
LICENSE ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Imene Guizani
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
app.py ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """LabLens web app (Gradio). Runs locally or as a Hugging Face Space.
2
+
3
+ Flow: upload -> read (deterministic extraction + flags) -> you check the table -> explain.
4
+ The check step is deliberate: OCR can misread a digit, and a human glance is the cheapest
5
+ safety net there is.
6
+ """
7
+ from __future__ import annotations
8
+
9
+ import html
10
+ from pathlib import Path
11
+
12
+ import gradio as gr
13
+
14
+ from lablens.catalog import load_catalog, match_test_name
15
+ from lablens.explain import explain, llm_available, render_markdown
16
+ from lablens.extract import extract
17
+ from lablens.flagger import flag_all
18
+ from lablens.i18n import CHECK_NOTE, DISCLAIMER, FLAG_LABEL, GENERIC_RANGE, LANGS, RTL, URGENT, t
19
+ from lablens.models import Flag, Result
20
+
21
+ SAMPLES = Path(__file__).parent / "samples"
22
+ HEADERS = ["Test", "Value", "Unit", "Range low", "Range high"]
23
+ LANG_CHOICES = [(v, k) for k, v in LANGS.items()]
24
+ catalog = load_catalog()
25
+
26
+ FLAG_COLOR = {
27
+ "normal": "ok", "low": "warn", "high": "warn", "critical_low": "crit", "critical_high": "crit",
28
+ "unknown": "muted", "check": "check",
29
+ }
30
+
31
+
32
+ def _num(x):
33
+ if x is None:
34
+ return None
35
+ s = str(x).strip().replace(",", ".")
36
+ if s in ("", "nan", "None", "-"):
37
+ return None
38
+ try:
39
+ return float(s)
40
+ except ValueError:
41
+ return None
42
+
43
+
44
+ def read_report(file, lang):
45
+ if file is None:
46
+ return gr.update(value=[]), "Upload a report or pick a sample first.", None
47
+ path = file if isinstance(file, str) else file.name
48
+ results, unmatched, detected, method = extract(path)
49
+ analysis = flag_all(results, method=method)
50
+ rows = [[catalog[r.test_key].display["en"], r.value, r.unit or "", r.ref_low, r.ref_high]
51
+ for r in analysis.results]
52
+ n_check = sum(r.flag == Flag.CHECK for r in analysis.results)
53
+ how = "the PDF's text" if method == "pdf-text" else "the photo (OCR)"
54
+ status = (f"Read **{len(rows)} results** from {how}. "
55
+ "Compare the table with your report, fix anything that's wrong, then press **Explain**.")
56
+ if n_check:
57
+ status += f"\n\n⚠️ {t(CHECK_NOTE, lang)}"
58
+ if unmatched:
59
+ status += f"\n\n<details><summary>{len(unmatched)} line(s) with numbers I couldn't match to a test</summary>\n\n" + \
60
+ "\n".join(f"- `{html.escape(u)}`" for u in unmatched[:15]) + "\n</details>"
61
+ return gr.update(value=rows), status, detected
62
+
63
+
64
+ def _rows_to_results(rows) -> list[Result]:
65
+ out = []
66
+ data = rows.values.tolist() if hasattr(rows, "values") else (rows or [])
67
+ for row in data:
68
+ if not row or not str(row[0]).strip():
69
+ continue
70
+ key, score = match_test_name(str(row[0]))
71
+ value = _num(row[1])
72
+ if key is None or value is None:
73
+ continue
74
+ lo, hi = _num(row[3]), _num(row[4])
75
+ out.append(Result(test_key=key, raw_name=str(row[0]), value=value, unit=str(row[2] or "") or None,
76
+ ref_low=lo, ref_high=hi,
77
+ ref_source="report" if (lo is not None or hi is not None) else "none",
78
+ match_score=score))
79
+ return out
80
+
81
+
82
+ def _results_table(analysis, lang) -> str:
83
+ rtl = ' dir="rtl"' if lang in RTL else ""
84
+ trs = []
85
+ for r in analysis.results:
86
+ td = catalog[r.test_key]
87
+ name = html.escape(td.display.get(lang) or td.display["en"])
88
+ if r.ref_low is not None and r.ref_high is not None:
89
+ rng = f"{r.ref_low:g} – {r.ref_high:g}"
90
+ elif r.ref_high is not None:
91
+ rng = f"&lt; {r.ref_high:g}"
92
+ elif r.ref_low is not None:
93
+ rng = f"&gt; {r.ref_low:g}"
94
+ else:
95
+ rng = "—"
96
+ if r.ref_source == "catalog":
97
+ rng += f'<div class="ll-note">{html.escape(t(GENERIC_RANGE, lang))}</div>'
98
+ label = html.escape(t(FLAG_LABEL[r.flag.value], lang))
99
+ trs.append(f'<tr><td>{name}</td><td class="ll-num">{r.value:g} <span class="ll-unit">{html.escape(r.unit or "")}</span></td>'
100
+ f'<td class="ll-num">{rng}</td><td><span class="ll-badge ll-{FLAG_COLOR[r.flag.value]}">{label}</span></td></tr>')
101
+ return f'<table class="ll-table"{rtl}><tbody>{"".join(trs)}</tbody></table>'
102
+
103
+
104
+ def run_explain(rows, lang):
105
+ results = _rows_to_results(rows)
106
+ if not results:
107
+ return "", "", "Nothing to explain yet: read a report first.", ""
108
+ analysis = flag_all(results)
109
+ rtl = ' dir="rtl"' if lang in RTL else ""
110
+ banner = ""
111
+ if analysis.urgent:
112
+ banner = f'<div class="ll-urgent"{rtl}>🚨 {_md_bold(t(URGENT, lang))}</div>'
113
+ if any(r.flag == Flag.CHECK for r in analysis.results):
114
+ banner += f'<div class="ll-check"{rtl}>⚠️ {html.escape(t(CHECK_NOTE, lang))}</div>'
115
+ exp, source = explain(analysis, lang)
116
+ md = render_markdown(analysis, exp, lang)
117
+ note = ("Explanation written by an open model, grounded in the LabLens knowledge base and checked by guardrails."
118
+ if source == "llm" else
119
+ "Explanation from the LabLens knowledge base (AI model not configured or unavailable).")
120
+ return banner, _results_table(analysis, lang), md, f"<small>{note} {html.escape(t(DISCLAIMER, lang))}</small>"
121
+
122
+
123
+ def _md_bold(s: str) -> str:
124
+ s = html.escape(s)
125
+ import re
126
+ return re.sub(r"\*\*(.+?)\*\*", r"<strong>\1</strong>", s)
127
+
128
+
129
+ def load_sample(name):
130
+ return str(SAMPLES / name)
131
+
132
+
133
+ CSS = """
134
+ :root { --ll-ink:#14213d; --ll-accent:#0f766e; }
135
+ .gradio-container { max-width: 1080px !important; margin: auto; }
136
+ #ll-hero h1 { font-size: 2.1rem; margin-bottom: .2rem; letter-spacing: -0.02em; }
137
+ #ll-hero p { font-size: 1.02rem; opacity: .85; margin-top: 0; }
138
+ .ll-pill { display:inline-block; padding:.18rem .6rem; border-radius:999px; font-size:.78rem;
139
+ background: color-mix(in srgb, var(--ll-accent) 14%, transparent); color: var(--ll-accent);
140
+ margin-right:.35rem; font-weight:600; }
141
+ .ll-table { width:100%; border-collapse:collapse; font-size:.95rem; }
142
+ .ll-table td { padding:.55rem .6rem; border-bottom:1px solid color-mix(in srgb, currentColor 12%, transparent); vertical-align: top; }
143
+ .ll-num { font-variant-numeric: tabular-nums; white-space: nowrap; }
144
+ .ll-unit { opacity:.65; font-size:.85em; }
145
+ .ll-note { font-size:.75rem; opacity:.65; white-space: normal; }
146
+ .ll-badge { padding:.15rem .55rem; border-radius:6px; font-size:.82rem; font-weight:600; white-space:nowrap; }
147
+ .ll-ok { background:#dcfce7; color:#14532d; }
148
+ .ll-warn { background:#fef3c7; color:#78350f; }
149
+ .ll-crit { background:#fee2e2; color:#7f1d1d; }
150
+ .ll-check { background:#e0e7ff; color:#312e81; padding:.7rem .9rem; border-radius:10px; margin:.4rem 0; }
151
+ .ll-muted { background:#e5e7eb; color:#374151; }
152
+ .ll-urgent { background:#fee2e2; color:#7f1d1d; border:2px solid #dc2626; padding:.9rem 1rem;
153
+ border-radius:12px; font-size:1.02rem; margin:.4rem 0; }
154
+ """
155
+
156
+ with gr.Blocks(title="LabLens") as demo:
157
+ gr.HTML(
158
+ '<div id="ll-hero"><h1>🔬 LabLens</h1>'
159
+ "<p>Understand your blood test in plain language — English, Français, العربية, or Tounsi.</p>"
160
+ '<span class="ll-pill">Flags by fixed medical rules, not AI</span>'
161
+ '<span class="ll-pill">Your image and name never reach the AI</span>'
162
+ '<span class="ll-pill">You check every value</span></div>'
163
+ )
164
+ with gr.Row():
165
+ with gr.Column(scale=5):
166
+ file = gr.File(label="Lab report (photo or PDF)", file_types=["image", ".pdf"], type="filepath")
167
+ with gr.Row():
168
+ s1 = gr.Button("Sample: English PDF", size="sm")
169
+ s2 = gr.Button("Sample: French photo", size="sm")
170
+ s3 = gr.Button("Sample: urgent value", size="sm")
171
+ lang = gr.Dropdown(LANG_CHOICES, value="en", label="Explain in")
172
+ read_btn = gr.Button("1 · Read report", variant="primary")
173
+ with gr.Column(scale=7):
174
+ status = gr.Markdown("Upload a report to begin.")
175
+ table = gr.Dataframe(headers=HEADERS, datatype=["str", "number", "str", "number", "number"],
176
+ interactive=True, label="What I read — edit anything that's wrong",
177
+ row_count=(0, "dynamic"), column_count=(5, "fixed"))
178
+ explain_btn = gr.Button("2 · Explain my results", variant="primary")
179
+
180
+ detected = gr.State(None)
181
+ banner = gr.HTML()
182
+ results_html = gr.HTML()
183
+ explanation = gr.Markdown()
184
+ footnote = gr.HTML(f"<small>{t(DISCLAIMER, 'en')} Synthetic sample reports only — fictional patients and labs.</small>")
185
+
186
+ s1.click(lambda: load_sample("sample_en_report.pdf"), outputs=file)
187
+ s2.click(lambda: (load_sample("sample_fr_photo.png"), "fr"), outputs=[file, lang])
188
+ s3.click(lambda: (load_sample("sample_fr_urgent.png"), "darija"), outputs=[file, lang])
189
+ read_btn.click(read_report, [file, lang], [table, status, detected])
190
+ explain_btn.click(run_explain, [table, lang], [banner, results_html, explanation, footnote])
191
+ lang.change(lambda l: gr.update(rtl=l in RTL), lang, explanation)
192
+
193
+ if __name__ == "__main__":
194
+ print("LLM configured:", llm_available())
195
+ demo.launch(css=CSS, theme=gr.themes.Soft(primary_hue="teal"))
packages.txt ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ tesseract-ocr
2
+ tesseract-ocr-eng
3
+ tesseract-ocr-fra
4
+ tesseract-ocr-ara
5
+ poppler-utils
6
+ fonts-dejavu-core
requirements.txt ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ gradio>=5.0
2
+ openai>=1.40
3
+ pydantic>=2.6
4
+ pyyaml>=6.0
5
+ rapidfuzz>=3.6
6
+ pdfplumber>=0.11
7
+ pytesseract>=0.3.10
8
+ opencv-python-headless>=4.9
9
+ numpy>=1.26
10
+ pillow>=10.0
11
+ reportlab>=4.0