FajarHidaa commited on
Commit
0886f37
·
verified ·
1 Parent(s): d000093

Delete src/core/pdf_processor.py

Browse files
Files changed (1) hide show
  1. src/core/pdf_processor.py +0 -283
src/core/pdf_processor.py DELETED
@@ -1,283 +0,0 @@
1
- """
2
- Human+ PDF Processor
3
- Extracts biomarkers from lab PDF using Gemini vision.
4
- Returns structured list of biomarker dicts ready for UI rendering.
5
- """
6
-
7
- import json
8
- import re
9
- from typing import Optional
10
- from core.ai_client import call_gemini, pdf_bytes_to_part, TEMP_EXTRACTION
11
-
12
-
13
- # ── Biomarker extraction prompt ──────────────────────────────────────
14
- _EXTRACTION_PROMPT = """
15
- Kamu adalah AI lab analyst dari Human+ Bali.
16
-
17
- Tugasmu: Ekstrak SEMUA biomarker dari hasil lab PDF ini.
18
-
19
- PENTING — Kembalikan HANYA valid JSON array, tanpa teks lain, tanpa markdown code block.
20
-
21
- Format setiap item:
22
- {
23
- "name": "Nama biomarker (gunakan nama standar, contoh: hs-CRP, Vitamin D, HbA1c)",
24
- "value": "Nilai numerik sebagai string (contoh: '2.4', '22', '5.6')",
25
- "unit": "Satuan (contoh: mg/L, ng/mL, %)",
26
- "reference_range": "Range referensi dari lab (contoh: 0.0-5.0 mg/L)",
27
- "lab_flag": "normal | high | low | critical_high | critical_low"
28
- }
29
-
30
- Prioritas biomarker (wajib cari jika ada):
31
- - Inflammation: hs-CRP, Homocysteine, Ferritin
32
- - Metabolic: HbA1c, Fasting Glucose, Fasting Insulin
33
- - Hormones: Total Testosterone, Free Testosterone, Vitamin D, DHEA-S, Cortisol, TSH
34
- - Nutrients: Vitamin B12, Magnesium, Zinc, Folate, Vitamin B6
35
- - Cardiovascular: ApoB, Triglycerides, HDL, LDL, Total Cholesterol
36
- - Blood: Hemoglobin, Hematocrit, WBC, Platelets
37
- - Liver: ALT, AST, GGT, Bilirubin
38
- - Kidney: Creatinine, eGFR, Uric Acid
39
-
40
- Jika nilai tidak terbaca atau tidak ada, SKIP biomarker tersebut.
41
- Kembalikan HANYA JSON array, tidak ada teks lain.
42
- """
43
-
44
- # ── Human+ optimal ranges (untuk status mapping) ─────────────────────
45
- _OPTIMAL_RANGES: dict[str, dict] = {
46
- "hs-crp": {"optimal": (None, 1.0), "warning": (1.0, 3.0)},
47
- "homocysteine": {"optimal": (None, 8.0), "warning": (8.0, 12.0)},
48
- "ferritin_m": {"optimal": (50, 150), "warning": (30, 50)}, # pria
49
- "ferritin_f": {"optimal": (30, 100), "warning": (15, 30)}, # wanita
50
- "hba1c": {"optimal": (4.8, 5.2), "warning": (5.2, 5.6)},
51
- "fasting glucose": {"optimal": (75, 86), "warning": (70, 100)},
52
- "fasting insulin": {"optimal": (None, 7.0), "warning": (7.0, 10.0)},
53
- "total testosterone": {"optimal": (600, None), "warning": (400, 600)},
54
- "vitamin d": {"optimal": (50, 80), "warning": (30, 50)},
55
- "dhea-s": {"optimal": None, "warning": None}, # age-dependent
56
- "vitamin b12": {"optimal": (500, None), "warning": (300, 500)},
57
- "magnesium": {"optimal": (2.2, None), "warning": (1.8, 2.2)},
58
- "zinc": {"optimal": (90, 120), "warning": (70, 90)},
59
- "folate": {"optimal": (10, None), "warning": (5, 10)},
60
- "apob": {"optimal": (None, 80), "warning": (80, 100)},
61
- "triglycerides": {"optimal": (None, 90), "warning": (90, 150)},
62
- "hdl": {"optimal": (60, None), "warning": (40, 60)},
63
- "tsh": {"optimal": (0.5, 2.0), "warning": (2.0, 4.0)},
64
- }
65
-
66
-
67
- class PDFProcessorError(Exception):
68
- """Raised when PDF extraction fails."""
69
- pass
70
-
71
-
72
- def extract_biomarkers(pdf_bytes: bytes) -> list[dict]:
73
- """
74
- Extract biomarkers dari PDF bytes menggunakan Gemini.
75
-
76
- Args:
77
- pdf_bytes: raw PDF file bytes
78
-
79
- Returns:
80
- list of biomarker dicts dengan keys:
81
- - name, value, unit, status, reference (Human+ optimal range)
82
- - raw_value, lab_flag, reference_range (dari lab asli)
83
-
84
- Raises:
85
- PDFProcessorError: jika extraction gagal
86
- """
87
- # 1. Call Gemini dengan PDF
88
- raw_json = _call_gemini_extraction(pdf_bytes)
89
-
90
- # 2. Parse JSON response
91
- raw_biomarkers = _parse_json_response(raw_json)
92
-
93
- # 3. Map ke Human+ status (optimal/warning/danger)
94
- enriched = [_enrich_biomarker(b) for b in raw_biomarkers]
95
-
96
- # 4. Filter yang invalid, sort by status priority
97
- valid = [b for b in enriched if b is not None]
98
- return _sort_by_priority(valid)
99
-
100
-
101
- def get_raw_extraction_text(pdf_bytes: bytes) -> str:
102
- """
103
- Return raw Gemini extraction text (untuk ditampilkan di expander debug).
104
- Useful untuk troubleshooting jika parsing gagal.
105
- """
106
- return _call_gemini_extraction(pdf_bytes)
107
-
108
-
109
- # ── Private helpers ──────────────────────────────────────────────────
110
-
111
- def _call_gemini_extraction(pdf_bytes: bytes) -> str:
112
- """Send PDF to Gemini and return raw text response."""
113
- try:
114
- pdf_part = pdf_bytes_to_part(pdf_bytes)
115
- response = call_gemini(
116
- contents=[_EXTRACTION_PROMPT, pdf_part],
117
- temperature=TEMP_EXTRACTION,
118
- )
119
- return response
120
- except Exception as e:
121
- raise PDFProcessorError(f"Gemini extraction gagal: {e}") from e
122
-
123
-
124
- def _parse_json_response(raw: str) -> list[dict]:
125
- """
126
- Parse JSON dari Gemini response.
127
- Gemini kadang return ```json ... ``` atau teks tambahan — kita strip dulu.
128
- """
129
- # Strip markdown code fences jika ada
130
- cleaned = re.sub(r"```(?:json)?", "", raw).strip().rstrip("```").strip()
131
-
132
- # Coba parse langsung
133
- try:
134
- data = json.loads(cleaned)
135
- if isinstance(data, list):
136
- return data
137
- # Kadang Gemini wrap dalam object
138
- if isinstance(data, dict):
139
- for key in ["biomarkers", "results", "data", "markers"]:
140
- if key in data and isinstance(data[key], list):
141
- return data[key]
142
- except json.JSONDecodeError:
143
- pass
144
-
145
- # Fallback: cari array JSON di dalam teks
146
- match = re.search(r"\[[\s\S]*\]", cleaned)
147
- if match:
148
- try:
149
- return json.loads(match.group())
150
- except json.JSONDecodeError:
151
- pass
152
-
153
- raise PDFProcessorError(
154
- "Gagal parse JSON dari Gemini. "
155
- "PDF mungkin tidak terbaca dengan baik atau format tidak standard."
156
- )
157
-
158
-
159
- def _enrich_biomarker(raw: dict) -> Optional[dict]:
160
- """
161
- Enrich raw biomarker dict dengan Human+ status mapping.
162
-
163
- Input keys: name, value, unit, reference_range, lab_flag
164
- Output adds: status ('optimal'|'warning'|'danger'|'neutral'), reference (Human+)
165
- """
166
- try:
167
- name = str(raw.get("name", "")).strip()
168
- value = str(raw.get("value", "")).strip()
169
- unit = str(raw.get("unit", "")).strip()
170
-
171
- if not name or not value:
172
- return None
173
-
174
- # Parse numeric value
175
- try:
176
- numeric_value = float(re.sub(r"[^\d.\-]", "", value))
177
- except (ValueError, TypeError):
178
- numeric_value = None
179
-
180
- # Map status
181
- status, human_plus_ref = _map_to_human_plus_status(
182
- name, numeric_value, raw.get("lab_flag", "normal")
183
- )
184
-
185
- # Gunakan Human+ target jika tersedia, fallback ke reference_range dari lab
186
- display_reference = human_plus_ref if human_plus_ref else raw.get("reference_range", "")
187
-
188
- return {
189
- # For UI rendering
190
- "name": name,
191
- "value": value,
192
- "unit": unit,
193
- "status": status,
194
- "reference": display_reference,
195
- # Raw lab data (untuk expander / debug)
196
- "raw_value": numeric_value,
197
- "lab_flag": raw.get("lab_flag", "normal"),
198
- "reference_range": raw.get("reference_range", ""),
199
- "human_plus_ref": human_plus_ref, # Human+ target khusus (bisa berbeda)
200
- }
201
- except Exception:
202
- return None
203
-
204
-
205
- def _map_to_human_plus_status(
206
- name: str,
207
- value: Optional[float],
208
- lab_flag: str,
209
- ) -> tuple[str, str]:
210
- """
211
- Map biomarker to Human+ optimal status.
212
-
213
- Returns:
214
- (status, human_plus_reference_string)
215
- status: 'optimal' | 'warning' | 'danger' | 'neutral'
216
- """
217
- # Normalisasi: hapus teks dalam kurung, lowercase
218
- # contoh: "hs-CRP (High-sensitivity C-Reactive Protein)" → "hs-crp"
219
- name_lower = re.sub(r"\s*\(.*?\)", "", name.lower()).strip()
220
-
221
- # Find matching range config
222
- range_cfg = None
223
- for key, cfg in _OPTIMAL_RANGES.items():
224
- if key == name_lower or key in name_lower or name_lower in key:
225
- range_cfg = cfg
226
- break
227
-
228
- # No config found — fallback ke lab_flag
229
- if range_cfg is None or value is None:
230
- if lab_flag in ("normal",):
231
- return "neutral", ""
232
- elif lab_flag in ("high", "low"):
233
- return "warning", ""
234
- elif lab_flag in ("critical_high", "critical_low"):
235
- return "danger", ""
236
- return "neutral", ""
237
-
238
- # Check optimal range
239
- optimal = range_cfg.get("optimal")
240
- warning = range_cfg.get("warning")
241
-
242
- if optimal:
243
- lo, hi = optimal
244
- in_optimal = (
245
- (lo is None or value >= lo) and
246
- (hi is None or value <= hi)
247
- )
248
- if in_optimal:
249
- ref_str = _format_range_str(lo, hi)
250
- return "optimal", ref_str
251
-
252
- if warning:
253
- lo, hi = warning
254
- in_warning = (
255
- (lo is None or value >= lo) and
256
- (hi is None or value <= hi)
257
- )
258
- if in_warning:
259
- opt_lo, opt_hi = (optimal or (None, None))
260
- ref_str = _format_range_str(opt_lo, opt_hi)
261
- return "warning", ref_str
262
-
263
- # Below all ranges or above all ranges → danger
264
- opt_lo, opt_hi = (optimal or (None, None))
265
- ref_str = _format_range_str(opt_lo, opt_hi)
266
- return "danger", ref_str
267
-
268
-
269
- def _format_range_str(lo: Optional[float], hi: Optional[float]) -> str:
270
- """Format optimal range sebagai human-readable string."""
271
- if lo is not None and hi is not None:
272
- return f"{lo}–{hi}"
273
- elif lo is not None:
274
- return f"> {lo}"
275
- elif hi is not None:
276
- return f"< {hi}"
277
- return ""
278
-
279
-
280
- def _sort_by_priority(biomarkers: list[dict]) -> list[dict]:
281
- """Sort: danger first, then warning, then optimal, then neutral."""
282
- priority = {"danger": 0, "warning": 1, "optimal": 2, "neutral": 3}
283
- return sorted(biomarkers, key=lambda b: priority.get(b["status"], 3))