{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.12.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[],"dockerImageVersionId":28755,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# =============================================================================\n# RSNA-ICA — analiza metatagów DICOM dla kohorty MRA\n# -----------------------------------------------------------------------------\n# Jedna komórka do Jupytera / Kaggle Notebooka. Wymaga podpiętych danych konkursu\n# \"RSNA Intracranial Aneurysm Detection\" (Add Input). Bez GPU i bez internetu.\n#\n# Co robi:\n# - czyta NAGŁÓWEK (bez pikseli) 1 pliku DICOM z każdej serii o etykiecie MRA,\n# - klasyfikuje serie: TOF / TOF po kontraście / CE-MRA / błąd etykiety,\n# - wypisuje rozkłady: producent, pole, parametry sekwencji, geometria,\n# - zapisuje tabelę 1 wiersz = 1 seria do mra_tags.csv.\n#\n# Uwagi:\n# - Philips zapisuje część serii jako Enhanced MR (1 plik = cały wolumen,\n# parametry w zagnieżdżonych sekwencjach). get_tag() przeszukuje cały\n# nagłówek, a TOF rozpoznajemy także po ImageType (...\\TOF\\...).\n# - Pusty ContrastBolusAgent (\"\") NIE oznacza kontrastu. Brak tagu też nie\n# dowodzi braku kontrastu (anonimizacja) -> \"bez udokumentowanego kontrastu\".\n# - Parametry sekwencji są zakładane jako stałe w obrębie serii (1 plik wystarcza).\n# =============================================================================\nimport os\nimport numpy as np\nimport pandas as pd\nimport pydicom\nfrom tqdm.auto import tqdm\n\nBASE = \"/kaggle/input/competitions/rsna-intracranial-aneurysm-detection\" # <- zmień, jeśli dane są gdzie indziej\nOUT_CSV = \"mra_tags.csv\"\n\npd.set_option(\"display.width\", 200, \"display.max_columns\", 30, \"display.max_rows\", 100)\n\n\n# ---------------------------------------------------------------- pomocnicze\ndef get_tag(ds, name):\n \"\"\"Wartość tagu z najwyższego poziomu albo z zagnieżdżonych sekwencji (Enhanced MR).\"\"\"\n if name in ds:\n return ds.get(name)\n for elem in ds.iterall():\n if elem.keyword == name:\n return elem.value\n return None\n\n\ndef as_str(v):\n if v is None:\n return None\n if isinstance(v, (list, tuple, pydicom.multival.MultiValue)):\n return \"\\\\\".join(str(x) for x in v)\n return str(v).strip()\n\n\ndef as_float(v):\n if isinstance(v, (list, tuple, pydicom.multival.MultiValue)):\n v = v[0] if len(v) else None\n try:\n return float(v)\n except (TypeError, ValueError):\n return np.nan\n\n\ndef vendor(m):\n m = str(m).upper()\n for key, name in [(\"SIEMENS\", \"Siemens\"), (\"GE\", \"GE\"), (\"PHILIPS\", \"Philips\"),\n (\"TOSHIBA\", \"Canon/Toshiba\"), (\"CANON\", \"Canon/Toshiba\"), (\"HITACHI\", \"Hitachi\")]:\n if key in m:\n return name\n return \"Inny/brak\"\n\n\ndef plane(iop):\n try:\n v = np.array([float(x) for x in iop])\n n = np.abs(np.cross(v[:3], v[3:]))\n return [\"strzałkowa\", \"czołowa\", \"osiowa\"][int(np.argmax(n))]\n except Exception:\n return None\n\n\n# ---------------------------------------------------------------- odczyt nagłówków\ntrain = pd.read_csv(f\"{BASE}/train.csv\")\nmra_uids = train.loc[train[\"Modality\"] == \"MRA\", \"SeriesInstanceUID\"].tolist()\n\nSTR_TAGS = [\"SeriesDescription\", \"ContrastBolusAgent\", \"Manufacturer\", \"ManufacturerModelName\",\n \"MRAcquisitionType\", \"ImageType\", \"ScanningSequence\", \"SequenceVariant\", \"AngioFlag\"]\nNUM_TAGS = [\"RepetitionTime\", \"EchoTime\", \"FlipAngle\", \"MagneticFieldStrength\",\n \"SliceThickness\", \"SpacingBetweenSlices\", \"PixelSpacing\", \"Rows\", \"Columns\", \"NumberOfFrames\"]\n\nrows = []\nfor uid in tqdm(mra_uids, desc=\"MRA\"):\n folder = f\"{BASE}/series/{uid}\"\n try:\n files = sorted(os.listdir(folder))\n ds = pydicom.dcmread(f\"{folder}/{files[0]}\", stop_before_pixels=True)\n r = {\"uid\": uid, \"n_files\": len(files), \"SOPClassUID\": str(ds.get(\"SOPClassUID\", \"\"))}\n for t in STR_TAGS:\n r[t] = as_str(get_tag(ds, t))\n for t in NUM_TAGS:\n r[t] = as_float(get_tag(ds, t))\n r[\"plaszczyzna\"] = plane(get_tag(ds, \"ImageOrientationPatient\"))\n rows.append(r)\n except Exception as e:\n rows.append({\"uid\": uid, \"error\": repr(e)[:200]})\n\ndf = pd.DataFrame(rows)\nif \"error\" not in df:\n df[\"error\"] = None\n\n# ---------------------------------------------------------------- pola pochodne\ndf[\"Vendor\"] = df[\"Manufacturer\"].apply(vendor)\ndf[\"Tesla\"] = df[\"MagneticFieldStrength\"].where(df[\"MagneticFieldStrength\"] < 100,\n df[\"MagneticFieldStrength\"] / 10000) # Gauss -> T\ndf[\"enhanced_mr\"] = df[\"SOPClassUID\"] == \"1.2.840.10008.5.1.4.1.1.4.1\"\ndf[\"n_przekrojow\"] = df[\"NumberOfFrames\"].fillna(df[\"n_files\"])\ndf[\"odstep_mm\"] = df[\"SpacingBetweenSlices\"].fillna(df[\"SliceThickness\"])\ndf[\"pokrycie_z_mm\"] = df[\"n_przekrojow\"] * df[\"odstep_mm\"]\ndf[\"nakladanie\"] = df[\"SliceThickness\"] > df[\"odstep_mm\"] + 0.01\n\nagent = df[\"ContrastBolusAgent\"].fillna(\"\").str.strip()\ndesc = df[\"SeriesDescription\"].fillna(\"\").str.upper()\nitype = df[\"ImageType\"].fillna(\"\").str.upper()\ndf[\"kontrast_tag\"] = agent != \"\"\ndf[\"post_w_opisie\"] = desc.str.contains(\"POST\")\n\n\ndef klasa(r):\n d = str(r[\"SeriesDescription\"]).upper()\n it = str(r[\"ImageType\"]).upper().split(\"\\\\\")\n if (\"T2\" in it) or (\"T2\" in d and \"TSE\" in d):\n return \"błąd etykiety (T2)\"\n if r[\"RepetitionTime\"] < 10:\n return \"CE-MRA\"\n is_tof = (r[\"RepetitionTime\"] >= 10) or (\"TOF\" in it)\n if is_tof and (r[\"kontrast_tag\"] or r[\"post_w_opisie\"]):\n return \"TOF po kontraście\"\n if is_tof:\n return \"TOF\"\n return \"nieustalone\"\n\n\nok = df[\"error\"].isna()\ndf.loc[ok, \"klasa\"] = df[ok].apply(klasa, axis=1)\ndf.loc[~ok, \"klasa\"] = \"błąd odczytu\"\ndf = df.merge(train[[\"SeriesInstanceUID\", \"Aneurysm Present\", \"PatientAge\", \"PatientSex\"]],\n left_on=\"uid\", right_on=\"SeriesInstanceUID\", how=\"left\").drop(columns=\"SeriesInstanceUID\")\ndf.to_csv(OUT_CSV, index=False)\n\n\n# ---------------------------------------------------------------- podsumowanie\ndef naglowek(t):\n print(f\"\\n{'=' * 80}\\n{t}\\n{'=' * 80}\")\n\n\ndef z_procentem(s):\n vc = s.value_counts(dropna=False)\n return pd.DataFrame({\"n\": vc, \"%\": (100 * vc / vc.sum()).round(1)})\n\n\ntof = df[df[\"klasa\"].isin([\"TOF\", \"TOF po kontraście\"])]\n\nnaglowek(f\"0. KOHORTA: {len(df)} serii MRA | błędy odczytu: {(~ok).sum()} | \"\n f\"Enhanced MR (1 plik = wolumen): {df['enhanced_mr'].sum()}\")\n\nnaglowek(\"1. KLASYFIKACJA SEKWENCJI\")\nprint(z_procentem(df[\"klasa\"]))\n\nnaglowek(\"2. ODSETEK TĘTNIAKÓW W KATEGORIACH\")\nprint(df.groupby(\"klasa\")[\"Aneurysm Present\"].agg(n=\"count\", tetniak=\"sum\", odsetek=\"mean\").round(3))\n\nnaglowek(\"3. KONTRAST (tag w nagłówku vs 'POST' w opisie) — cała kohorta\")\nprint(pd.crosstab(df[\"kontrast_tag\"], df[\"post_w_opisie\"], margins=True))\n\nnaglowek(\"4. PRODUCENT\")\nprint(pd.crosstab(df[\"Vendor\"], df[\"klasa\"], margins=True))\n\nnaglowek(\"5. NAJCZĘSTSZE MODELE SKANERÓW (top 15)\")\nprint((df[\"Vendor\"] + \" | \" + df[\"ManufacturerModelName\"].fillna(\"brak\")).value_counts().head(15))\n\nnaglowek(\"6. POLE MAGNETYCZNE [T]\")\nprint(pd.crosstab(df[\"Tesla\"].round(1).fillna(-1).rename(\"Tesla (-1 = brak)\"), df[\"klasa\"], margins=True))\nprint()\nprint(pd.crosstab(df[\"Vendor\"], df[\"Tesla\"].round(1).fillna(-1).rename(\"Tesla\"), margins=True))\n\nnaglowek(\"7. PARAMETRY SEKWENCJI wg klasy (mediana [min–max])\")\ndef med_zakres(x):\n x = x.dropna()\n return f\"{x.median():.2f} [{x.min():.2f}–{x.max():.2f}] n={len(x)}\" if len(x) else \"brak\"\nprint(df.groupby(\"klasa\")[[\"RepetitionTime\", \"EchoTime\", \"FlipAngle\"]].agg(med_zakres))\n\nnaglowek(f\"8. GEOMETRIA — tylko TOF (n={len(tof)}), percentyle\")\ngeo_cols = [\"SliceThickness\", \"SpacingBetweenSlices\", \"odstep_mm\", \"PixelSpacing\",\n \"Rows\", \"Columns\", \"n_przekrojow\", \"pokrycie_z_mm\"]\nprint(tof[geo_cols].describe(percentiles=[.05, .25, .5, .75, .95]).T.round(2))\n\nnaglowek(\"9. GEOMETRIA TOF — przedziały\")\nprint(\"Grubość warstwy [mm]:\")\nprint(z_procentem(pd.cut(tof[\"SliceThickness\"], [0, 0.5, 0.7, 1.0, 1.5, 100])).sort_index())\nprint(\"\\nOdstęp między przekrojami [mm]:\")\nprint(z_procentem(pd.cut(tof[\"odstep_mm\"], [0, 0.3, 0.5, 0.7, 1.0, 100])).sort_index())\nprint(\"\\nRozmiar piksela w płaszczyźnie [mm]:\")\nprint(z_procentem(pd.cut(tof[\"PixelSpacing\"], [0, 0.3, 0.4, 0.5, 0.7, 100])).sort_index())\nprint(\"\\nLiczba przekrojów w serii:\")\nprint(z_procentem(pd.cut(tof[\"n_przekrojow\"], [0, 60, 100, 150, 200, 300, 10000])).sort_index())\nprint(\"\\nNakładanie przekrojów (odstęp < grubość):\")\nprint(z_procentem(tof[\"nakladanie\"]))\n\nnaglowek(\"10. GRUBOŚĆ WARSTWY TOF wg producenta i pola (mediana [mm])\")\nprint(tof.pivot_table(index=\"Vendor\", columns=tof[\"Tesla\"].round(1), values=\"SliceThickness\",\n aggfunc=\"median\").round(2))\n\nnaglowek(\"11. PŁASZCZYZNA × 2D/3D (TOF)\")\nprint(pd.crosstab(tof[\"plaszczyzna\"].fillna(\"brak\"), tof[\"MRAcquisitionType\"].fillna(\"brak\"), margins=True))\n\nnaglowek(\"12. ImageType (TOF) — czy są MIP-y / rekonstrukcje (DERIVED)?\")\nprint(tof[\"ImageType\"].fillna(\"brak\").value_counts().head(15))\nprint(\"\\nSerie DERIVED lub z 'MIP' w ImageType/opisie:\",\n (tof[\"ImageType\"].fillna(\"\").str.upper().str.contains(\"DERIVED|MIP\") |\n tof[\"SeriesDescription\"].fillna(\"\").str.upper().str.contains(\"MIP\")).sum())\n\nnaglowek(\"13. NAJCZĘSTSZE SeriesDescription (top 20)\")\nprint(desc.value_counts().head(20))\n\nnaglowek(\"14. SERIE SPOZA TOF (do ręcznego przejrzenia)\")\nprint(df.loc[~df[\"klasa\"].isin([\"TOF\", \"TOF po kontraście\"]),\n [\"klasa\", \"SeriesDescription\", \"ContrastBolusAgent\", \"RepetitionTime\",\n \"EchoTime\", \"FlipAngle\", \"Vendor\", \"Tesla\"]].to_string())\n\nprint(f\"\\nZapisano: {OUT_CSV} ({len(df)} wierszy, 1 wiersz = 1 seria)\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T13:17:48.83238Z","iopub.execute_input":"2026-09-24T13:17:48.832965Z","iopub.status.idle":"2026-09-24T13:17:59.57517Z","shell.execute_reply.started":"2026-09-24T13:17:48.832933Z","shell.execute_reply":"2026-09-24T13:17:59.574327Z"}},"outputs":[{"output_type":"display_data","data":{"text/plain":"MRA: 0%| | 0/1252 [00:00\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
uidn_filesSeriesDescriptionContrastBolusAgentRepetitionTimeEchoTimeFlipAngleManufacturerMagneticFieldStrength
01.2.826.0.1.3680043.8.498.10004044428023505108...188AX COWNone24725SIEMENS1.5
11.2.826.0.1.3680043.8.498.10004684224894397679...1473D TOF Brain*None232.720GE MEDICAL SYSTEMS1.5
21.2.826.0.1.3680043.8.498.10009383108068795488...224AX 3D TOF WHOLE HEADNone213.4220Siemens Healthineers3
31.2.826.0.1.3680043.8.498.10012790035410518400...1MRANoneNoneNoneNonePhilips Medical Systems1.5
41.2.826.0.1.3680043.8.498.10022688097731894079...178ax tof cowNone213.4318SIEMENS3
51.2.826.0.1.3680043.8.498.10023411164590664678...205TOF_FL3D_TRANone44.883.515Siemens Healthineers3
61.2.826.0.1.3680043.8.498.10035782880104673269...1MRANoneNoneNoneNonePhilips Medical Systems1.5
71.2.826.0.1.3680043.8.498.10037266473301611864...201AX COW PRENone223.618SIEMENS3
81.2.826.0.1.3680043.8.498.10040419508532196461...188ax 3dtofNone20320GE MEDICAL SYSTEMS1.5
91.2.826.0.1.3680043.8.498.10042423585566957032...116TOF_3D MRA .7None227.1525SIEMENS1.5
\n"},"metadata":{}}],"execution_count":4},{"cell_type":"code","source":"mra_meta.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T12:57:18.14211Z","iopub.execute_input":"2026-09-24T12:57:18.14241Z","iopub.status.idle":"2026-09-24T12:57:18.154778Z","shell.execute_reply.started":"2026-09-24T12:57:18.142386Z","shell.execute_reply":"2026-09-24T12:57:18.153817Z"}},"outputs":[{"execution_count":5,"output_type":"execute_result","data":{"text/plain":" uid n_files \\\n0 1.2.826.0.1.3680043.8.498.10004044428023505108... 188 \n1 1.2.826.0.1.3680043.8.498.10004684224894397679... 147 \n2 1.2.826.0.1.3680043.8.498.10009383108068795488... 224 \n3 1.2.826.0.1.3680043.8.498.10012790035410518400... 1 \n4 1.2.826.0.1.3680043.8.498.10022688097731894079... 178 \n\n SeriesDescription ContrastBolusAgent RepetitionTime EchoTime FlipAngle \\\n0 AX COW None 24 7 25 \n1 3D TOF Brain* None 23 2.7 20 \n2 AX 3D TOF WHOLE HEAD None 21 3.42 20 \n3 MRA None None None None \n4 ax tof cow None 21 3.43 18 \n\n Manufacturer MagneticFieldStrength \n0 SIEMENS 1.5 \n1 GE MEDICAL SYSTEMS 1.5 \n2 Siemens Healthineers 3 \n3 Philips Medical Systems 1.5 \n4 SIEMENS 3 ","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
uidn_filesSeriesDescriptionContrastBolusAgentRepetitionTimeEchoTimeFlipAngleManufacturerMagneticFieldStrength
01.2.826.0.1.3680043.8.498.10004044428023505108...188AX COWNone24725SIEMENS1.5
11.2.826.0.1.3680043.8.498.10004684224894397679...1473D TOF Brain*None232.720GE MEDICAL SYSTEMS1.5
21.2.826.0.1.3680043.8.498.10009383108068795488...224AX 3D TOF WHOLE HEADNone213.4220Siemens Healthineers3
31.2.826.0.1.3680043.8.498.10012790035410518400...1MRANoneNoneNoneNonePhilips Medical Systems1.5
41.2.826.0.1.3680043.8.498.10022688097731894079...178ax tof cowNone213.4318SIEMENS3
\n
"},"metadata":{}}],"execution_count":5},{"cell_type":"code","source":"for c in [\"RepetitionTime\", \"EchoTime\", \"FlipAngle\"]:\n mra_meta[c] = pd.to_numeric(mra_meta[c], errors=\"coerce\")\n\nprint(\"Kontrast wpisany w nagłówku:\", mra_meta[\"ContrastBolusAgent\"].notna().sum(), \"/\", len(mra_meta))\nprint(\"\\nTR < 10 ms (podejrzenie CE-MRA):\", (mra_meta[\"RepetitionTime\"] < 10).sum())\nprint(\"TR >= 10 ms (podejrzenie TOF): \", (mra_meta[\"RepetitionTime\"] >= 10).sum())\nprint(\"\\nNajczęstsze opisy serii:\")\nprint(mra_meta[\"SeriesDescription\"].str.upper().value_counts().head(20))\nprint(\"\\nProducenci:\")\nprint(mra_meta[\"Manufacturer\"].value_counts())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T12:57:18.155931Z","iopub.execute_input":"2026-09-24T12:57:18.156379Z","iopub.status.idle":"2026-09-24T12:57:18.181112Z","shell.execute_reply.started":"2026-09-24T12:57:18.156347Z","shell.execute_reply":"2026-09-24T12:57:18.180298Z"}},"outputs":[{"name":"stdout","text":"Kontrast wpisany w nagłówku: 97 / 1252\n\nTR < 10 ms (podejrzenie CE-MRA): 13\nTR >= 10 ms (podejrzenie TOF): 1082\n\nNajczęstsze opisy serii:\nSeriesDescription\nMRA 156\nTOF_3D_MULTI-SLAB 109\nAX COW 78\nTOF_FL3D_TRA 59\nAX TOF COW 42\nAX TOF COW 0.6 41\nAX 3D TOF BRAIN 31\nAX 3DTOF 26\nH-AX 3D TOF-COW 25\nBRAIN AX 3D MRA_TOF 23\nTOF_3D_MULTI-SLAB_HEAD 22\nTOF_3D_5X_MULTI-SLAB 384 17\nTOF_CS_ACC5.0 17\n3D TOF MRA NEW 17\nCOW 15\nAX 3DTOF 3SLAB 13\nAX 3D TOF 13\nAX TOF 3D - MRA COW 13\nAX 3DTOF 2SLAB 12\nAX 3D TOF SLAB 12\nName: count, dtype: int64\n\nProducenci:\nManufacturer\nSIEMENS 381\nGE MEDICAL SYSTEMS 298\nSiemens Healthineers 231\nPhilips Medical Systems 179\nTOSHIBA 48\nSiemens 45\nPhilips 30\nCANON_MEC 17\nPhilips Healthcare 15\nTOSHIBA_MEC 5\nSiemens HealthCare GmbH 2\nHitachi, Ltd. 1\nName: count, dtype: int64\n","output_type":"stream"}],"execution_count":6},{"cell_type":"code","source":"# 1. Ujednolicenie producentów\ndef vendor(m):\n m = str(m).upper()\n for key, name in [(\"SIEMENS\", \"Siemens\"), (\"GE\", \"GE\"), (\"PHILIPS\", \"Philips\"),\n (\"TOSHIBA\", \"Canon/Toshiba\"), (\"CANON\", \"Canon/Toshiba\"), (\"HITACHI\", \"Hitachi\")]:\n if key in m:\n return name\n return \"Inny/brak\"\n\nmra_meta[\"Vendor\"] = mra_meta[\"Manufacturer\"].apply(vendor)\nmra_meta[\"contrast\"] = mra_meta[\"ContrastBolusAgent\"].notna()\nprint(mra_meta[\"Vendor\"].value_counts(), \"\\n\")\n\n# 2. Te 157 serii bez TR - co to jest?\nno_tr = mra_meta[mra_meta[\"RepetitionTime\"].isna()]\nprint(\"Serie bez TR:\", len(no_tr))\nif \"error\" in mra_meta:\n print(\" w tym błędy odczytu:\", no_tr[\"error\"].notna().sum())\nprint(no_tr[[\"SeriesDescription\", \"Vendor\", \"n_files\"]].value_counts().head(15), \"\\n\")\n\n# 3. Kontrast w nagłówku vs TR\nmra_meta[\"TR_grupa\"] = pd.cut(mra_meta[\"RepetitionTime\"], [0, 10, 1000], labels=[\"TR<10\", \"TR>=10\"])\nprint(pd.crosstab(mra_meta[\"TR_grupa\"].astype(str), mra_meta[\"contrast\"], margins=True), \"\\n\")\n\n# 4. Szczegóły serii z kontrastem\ncols = [\"SeriesDescription\", \"ContrastBolusAgent\", \"RepetitionTime\", \"EchoTime\", \"FlipAngle\", \"Vendor\"]\nprint(mra_meta.loc[mra_meta[\"contrast\"], cols].to_string(max_rows=40))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T12:58:27.617051Z","iopub.execute_input":"2026-09-24T12:58:27.618146Z","iopub.status.idle":"2026-09-24T12:58:27.69409Z","shell.execute_reply.started":"2026-09-24T12:58:27.618096Z","shell.execute_reply":"2026-09-24T12:58:27.693176Z"}},"outputs":[{"name":"stdout","text":"Vendor\nSiemens 659\nGE 298\nPhilips 224\nCanon/Toshiba 70\nHitachi 1\nName: count, dtype: int64 \n\nSerie bez TR: 157\nSeriesDescription Vendor n_files\nMRA Philips 1 156\nT2 TSE Philips 1 1\nName: count, dtype: int64 \n\ncontrast False True All\nTR_grupa \nTR<10 5 8 13\nTR>=10 993 89 1082\nnan 157 0 157\nAll 1155 97 1252 \n\n SeriesDescription ContrastBolusAgent RepetitionTime EchoTime FlipAngle Vendor\n18 TOF_FL3D_TRA_CS7.2 20.350 3.690 20.0 Siemens\n50 AX 3D TOF COW POST PROHANCE 21.000 3.420 20.0 Siemens\n64 AX COW POST NO SUPERIOR SATBAND PER UR 17 ML PROHANCE 22.000 3.600 18.0 Siemens\n96 Brain Ax 3D MRA_TOF 43.940 2.880 22.0 Siemens\n101 AX COW POST 12 ML PROHANCE 22.000 3.600 18.0 Siemens\n104 AX 3D TOF POST 20 ML PROHANCE 22.000 3.430 18.0 Siemens\n111 AX 3D TOF COW POST PROHANCE 21.000 3.420 20.0 Siemens\n138 AX COW POST 19 ML MULTIHANCE 22.000 3.600 18.0 Siemens\n144 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n165 cemra post Multihance 3.420 1.230 23.0 Siemens\n176 AX 3D TOF COW POST PROHANCE 21.000 3.420 18.0 Siemens\n177 AX COW POST 20 ML PROHANCE 22.000 3.600 18.0 Siemens\n189 AX 3D COW 23.000 3.455 18.0 Philips\n204 h-Ax.s3DI_COW 23.000 3.455 18.0 Philips\n207 AX COW POST NO SUPERIOR SATBAND PER UR 18 ML PROHANCE 22.000 3.600 18.0 Siemens\n231 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n258 Ax.s3DI_COW 23.000 3.455 18.0 Philips\n284 AX COW POST NO SUPERIOR SATBAND PER UR 12 mL MULTIHANCE 22.000 3.600 18.0 Siemens\n308 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n333 AX 3D TOF COW POST PROHANCE 21.000 3.420 20.0 Siemens\n... ... ... ... ... ... ...\n1032 AX 3D COW 23.000 3.455 18.0 Philips\n1038 TOF_CS_WHOLE_BRAIN 23.370 7.120 20.0 Siemens\n1039 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n1065 3D Ax TOF SPGR FS HyperSense YES 10 22.000 2.500 15.0 GE\n1076 cemra post multihance 3.420 1.230 23.0 Siemens\n1078 AX COW POST WB 21 ML MULTIHANCE 22.000 3.600 18.0 Siemens\n1101 AX 3D TOF POST 10ML PROHANCE 21.000 3.430 18.0 Siemens\n1130 Ax.s3DI_COW 23.000 3.455 18.0 Philips\n1138 TOF_FL3D_TRA_CS7.2 20.350 3.690 20.0 Siemens\n1140 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n1148 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n1159 TOF_MRA_3D_AX_ 25.000 7.000 25.0 Siemens\n1169 AX 3D TOF COW POST PROHANCE 21.000 3.420 20.0 Siemens\n1187 AX COW POST 10 ML PROHANCE 29.000 7.000 25.0 Siemens\n1188 TOF_MRA_3D_AX_ 25.000 7.000 25.0 Siemens\n1194 AX 3D TOF COW POST PROHANCE 21.000 3.420 20.0 Siemens\n1207 TOF_CS_ACC5.0 23.290 7.120 20.0 Siemens\n1229 tof_fl3d_tra fast 2019 new PROHANCE 21.000 3.420 20.0 Siemens\n1237 tof_fl3d_tra PROHANCE 21.000 3.420 20.0 Siemens\n1243 None DeIdentified 4.056 1.468 25.0 GE\n","output_type":"stream"}],"execution_count":7},{"cell_type":"code","source":"# A. Poprawiona flaga kontrastu: tylko niepusty tekst\nagent = mra_meta[\"ContrastBolusAgent\"].fillna(\"\").str.strip()\nmra_meta[\"contrast\"] = agent != \"\"\nmra_meta[\"post_w_opisie\"] = mra_meta[\"SeriesDescription\"].fillna(\"\").str.upper().str.contains(\"POST\")\nprint(pd.crosstab(mra_meta[\"TR_grupa\"].astype(str),\n [mra_meta[\"contrast\"], mra_meta[\"post_w_opisie\"]], margins=True), \"\\n\")\n\n# B. Wszystkie 13 serii z TR < 10 ms\ncols = [\"SeriesDescription\", \"ContrastBolusAgent\", \"RepetitionTime\", \"EchoTime\", \"FlipAngle\", \"Vendor\", \"n_files\"]\nprint(mra_meta.loc[mra_meta[\"RepetitionTime\"] < 10, cols].to_string(), \"\\n\")\n\n# C. Zajrzyjmy do jednej serii Philips z 1 plikiem\nuid = mra_meta.loc[mra_meta[\"RepetitionTime\"].isna() & (mra_meta[\"SeriesDescription\"] == \"MRA\"), \"uid\"].iloc[0]\nfolder = f\"{BASE}/series/{uid}\"\nds = pydicom.dcmread(f\"{folder}/{os.listdir(folder)[0]}\", stop_before_pixels=True)\nprint(\"SOPClassUID:\", ds.get(\"SOPClassUID\"))\nprint(\"NumberOfFrames:\", ds.get(\"NumberOfFrames\"))\nprint(\"ImageType:\", ds.get(\"ImageType\"))\nprint(\"Rows x Columns:\", ds.get(\"Rows\"), \"x\", ds.get(\"Columns\"))\nprint(\"Ma SharedFunctionalGroups:\", \"SharedFunctionalGroupsSequence\" in ds)\nprint(\"Ma PerFrameFunctionalGroups:\", \"PerFrameFunctionalGroupsSequence\" in ds)\nprint(\"\\nWszystkie tagi najwyższego poziomu:\")\nprint([e.keyword or str(e.tag) for e in ds])","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T12:58:32.637095Z","iopub.execute_input":"2026-09-24T12:58:32.638153Z","iopub.status.idle":"2026-09-24T12:58:32.698961Z","shell.execute_reply.started":"2026-09-24T12:58:32.638116Z","shell.execute_reply":"2026-09-24T12:58:32.69808Z"}},"outputs":[{"name":"stdout","text":"contrast False True All\npost_w_opisie False True False True \nTR_grupa \nTR<10 5 0 2 6 13\nTR>=10 1041 1 11 29 1082\nnan 157 0 0 0 157\nAll 1203 1 13 35 1252 \n\n SeriesDescription ContrastBolusAgent RepetitionTime EchoTime FlipAngle Vendor n_files\n42 cemra pre None 3.4200 1.230 23.0 Siemens 104\n165 cemra post Multihance 3.4200 1.230 23.0 Siemens 120\n245 None None 5.4798 1.872 35.0 Philips 160\n473 cemra post_TTC=1.0s MULTIHANCE 3.4700 1.330 17.0 Siemens 104\n478 cemra pre_TTC=1.0s None 3.4700 1.330 17.0 Siemens 96\n479 cemra post_TTC=1.0s MULTIHANCE 3.4700 1.330 17.0 Siemens 96\n647 None None 3.8560 1.396 25.0 GE 248\n773 None DeIdentified 3.2000 1.120 25.0 Siemens 72\n872 cemra post_TTC=1.0s MULTIHANCE 3.4700 1.330 17.0 Siemens 96\n897 cemra post multihance 3.4200 1.230 23.0 Siemens 104\n1076 cemra post multihance 3.4200 1.230 23.0 Siemens 104\n1222 Ph1/Cor ceMRA FT elliptic None 5.0360 1.776 25.0 GE 116\n1243 None DeIdentified 4.0560 1.468 25.0 GE 100 \n\nSOPClassUID: 1.2.840.10008.5.1.4.1.1.4.1\nNumberOfFrames: 150\nImageType: ['ORIGINAL', 'PRIMARY', 'TOF', 'NONE']\nRows x Columns: 528 x 528\nMa SharedFunctionalGroups: True\nMa PerFrameFunctionalGroups: True\n\nWszystkie tagi najwyższego poziomu:\n['SpecificCharacterSet', 'ImageType', 'SOPClassUID', 'SOPInstanceUID', 'StudyDate', 'SeriesDate', 'ContentDate', 'AcquisitionDateTime', 'StudyTime', 'SeriesTime', 'ContentTime', 'AccessionNumber', 'Modality', 'Manufacturer', 'StudyDescription', 'SeriesDescription', 'ManufacturerModelName', 'PatientName', 'PatientID', 'PatientSex', 'PatientWeight', 'PatientIdentityRemoved', 'DeidentificationMethod', 'DeidentificationMethodCodeSequence', '(0013,0010)', '(0013,1001)', '(0013,1003)', 'BodyPartExamined', 'MRAcquisitionType', 'MagneticFieldStrength', 'SpacingBetweenSlices', 'PixelBandwidth', 'SoftwareVersions', 'PatientPosition', 'ContentQualification', 'PulseSequenceName', 'EchoPulseSequence', 'MultiPlanarExcitation', 'PhaseContrast', 'TimeOfFlightContrast', 'Spoiling', 'SteadyStatePulseSequence', 'EchoPlanarPulseSequence', 'MagnetizationTransfer', 'T2Preparation', 'BloodSignalNulling', 'SaturationRecovery', 'SpectrallySelectedSuppression', 'SpectrallySelectedExcitation', 'SpatialPresaturation', 'Tagging', 'OversamplingPhase', 'GeometryOfKSpaceTraversal', 'SegmentedKSpaceTraversal', 'RectilinearPhaseEncodeReordering', 'TagThickness', 'PartialFourierDirection', 'CardiacSynchronizationTechnique', 'TransmitCoilType', 'ChemicalShiftReference', 'MRAcquisitionFrequencyEncodingSteps', 'Decoupling', 'KSpaceFiltering', 'ParallelReductionFactorInPlane', 'AcquisitionDuration', 'ParallelAcquisition', 'ParallelAcquisitionTechnique', 'PartialFourier', 'VelocityEncodingDirection', 'VelocityEncodingMinimumValue', 'NumberOfKSpaceTrajectories', 'CoverageOfKSpace', 'ResonantNucleus', 'FrequencyCorrection', 'ParallelReductionFactorOutOfPlane', 'ParallelReductionFactorSecondInPlane', 'RespiratoryMotionCompensationTechnique', 'BulkMotionCompensationTechnique', 'ApplicableSafetyStandardAgency', 'SpecificAbsorptionRateDefinition', 'GradientOutputType', 'SpecificAbsorptionRateValue', 'GradientOutput', 'FlowCompensationDirection', 'WaterReferencedPhaseCorrection', 'MRSpectroscopyAcquisitionType', 'MRAcquisitionPhaseEncodingStepsInPlane', 'RFEchoTrainLength', 'GradientEchoTrainLength', 'StudyInstanceUID', 'SeriesInstanceUID', 'StudyID', 'SeriesNumber', 'AcquisitionNumber', 'InstanceNumber', 'FrameOfReferenceUID', 'PositionReferenceIndicator', 'FrameLaterality', 'DimensionOrganizationSequence', 'DimensionIndexSequence', 'RespiratoryIntervalTime', 'NominalRespiratoryTriggerDelayTime', 'SamplesPerPixel', 'PhotometricInterpretation', 'NumberOfFrames', 'Rows', 'Columns', 'BitsAllocated', 'BitsStored', 'HighBit', 'PixelRepresentation', 'BurnedInAnnotation', 'LossyImageCompression', 'LUTExplanation', 'DataPointRows', 'DataPointColumns', 'SharedFunctionalGroupsSequence', 'PerFrameFunctionalGroupsSequence']\n","output_type":"stream"}],"execution_count":8},{"cell_type":"code","source":"# Sprawdzenie ImageType dla 157 serii bez TR\ndef image_type(uid):\n folder = f\"{BASE}/series/{uid}\"\n ds = pydicom.dcmread(f\"{folder}/{os.listdir(folder)[0]}\", stop_before_pixels=True)\n return \"\\\\\".join(ds.get(\"ImageType\", []))\n\nbez_tr = mra_meta[\"RepetitionTime\"].isna()\nmra_meta.loc[bez_tr, \"ImageType\"] = mra_meta.loc[bez_tr, \"uid\"].apply(image_type)\nprint(mra_meta.loc[bez_tr, [\"SeriesDescription\", \"ImageType\"]].value_counts(), \"\\n\")\n\n# Ostateczna klasyfikacja\ndef klasa(r):\n opis = str(r[\"SeriesDescription\"]).upper()\n if \"T2\" in opis and \"TSE\" in opis:\n return \"błąd etykiety (T2)\"\n if r[\"RepetitionTime\"] < 10:\n return \"CE-MRA\"\n if r[\"contrast\"] or r[\"post_w_opisie\"]:\n return \"TOF po kontraście\"\n if r[\"RepetitionTime\"] >= 10 or \"TOF\" in str(r.get(\"ImageType\", \"\")):\n return \"TOF\"\n return \"nieustalone\"\n\nmra_meta[\"klasa\"] = mra_meta.apply(klasa, axis=1)\nprint(mra_meta[\"klasa\"].value_counts(), \"\\n\")\n\n# Pole magnetyczne i producent per klasa\nmra_meta[\"Tesla\"] = pd.to_numeric(mra_meta[\"MagneticFieldStrength\"], errors=\"coerce\")\nmra_meta.loc[mra_meta[\"Tesla\"] > 100, \"Tesla\"] /= 10000 # Gauss -> Tesla\nprint(pd.crosstab(mra_meta[\"Tesla\"].round(1), mra_meta[\"klasa\"], margins=True), \"\\n\")\nprint(pd.crosstab(mra_meta[\"Vendor\"], mra_meta[\"klasa\"], margins=True))\n\nmra_meta.to_csv(\"mra_meta_klasy.csv\", index=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T13:02:43.287248Z","iopub.execute_input":"2026-09-24T13:02:43.287687Z","iopub.status.idle":"2026-09-24T13:02:44.056209Z","shell.execute_reply.started":"2026-09-24T13:02:43.287655Z","shell.execute_reply":"2026-09-24T13:02:44.055332Z"}},"outputs":[{"name":"stdout","text":"SeriesDescription ImageType \nMRA ORIGINAL\\PRIMARY\\TOF\\NONE 156\nT2 TSE ORIGINAL\\PRIMARY\\T2\\NONE 1\nName: count, dtype: int64 \n\nklasa\nTOF 1197\nTOF po kontraście 41\nCE-MRA 13\nbłąd etykiety (T2) 1\nName: count, dtype: int64 \n\nklasa CE-MRA TOF TOF po kontraście błąd etykiety (T2) All\nTesla \n1.2 0 1 0 0 1\n1.5 8 694 5 1 708\n3.0 5 497 36 0 538\nAll 13 1192 41 1 1247 \n\nklasa CE-MRA TOF TOF po kontraście błąd etykiety (T2) All\nVendor \nCanon/Toshiba 0 70 0 0 70\nGE 3 289 6 0 298\nHitachi 0 1 0 0 1\nPhilips 1 222 0 1 224\nSiemens 9 615 35 0 659\nAll 13 1197 41 1 1252\n","output_type":"stream"}],"execution_count":9},{"cell_type":"code","source":"import numpy as np\n\nTAGS_MR = [\"SeriesDescription\", \"ContrastBolusAgent\", \"RepetitionTime\", \"EchoTime\",\n \"InversionTime\", \"FlipAngle\", \"ImageType\", \"ScanningSequence\",\n \"MRAcquisitionType\", \"Manufacturer\", \"MagneticFieldStrength\",\n \"SliceThickness\", \"ImageOrientationPatient\"]\n\ndef plaszczyzna(iop):\n try:\n v = np.array([float(x) for x in iop])\n n = np.abs(np.cross(v[:3], v[3:]))\n return [\"strzałkowa\", \"czołowa\", \"osiowa\"][int(np.argmax(n))]\n except Exception:\n return None\n\nrows = []\nmr_uids = train.loc[train[\"Modality\"].isin([\"MRI T2\", \"MRI T1post\"]), [\"SeriesInstanceUID\", \"Modality\"]]\nfor uid, mod in tqdm(mr_uids.values):\n folder = f\"{BASE}/series/{uid}\"\n try:\n files = os.listdir(folder)\n ds = pydicom.dcmread(f\"{folder}/{sorted(files)[0]}\", stop_before_pixels=True)\n row = {\"uid\": uid, \"etykieta\": mod, \"n_files\": len(files)}\n for t in TAGS_MR:\n v = get_tag(ds, t)\n if t == \"ImageOrientationPatient\":\n row[\"plaszczyzna\"] = plaszczyzna(v) if v is not None else None\n elif t in (\"ImageType\", \"ScanningSequence\") and v is not None and not isinstance(v, str):\n row[t] = \"\\\\\".join(str(x) for x in v)\n else:\n row[t] = str(v) if v is not None else None\n rows.append(row)\n except Exception as e:\n rows.append({\"uid\": uid, \"etykieta\": mod, \"error\": str(e)})\n\nmr_meta = pd.DataFrame(rows)\nfor c in [\"RepetitionTime\", \"EchoTime\", \"InversionTime\", \"FlipAngle\", \"SliceThickness\"]:\n mr_meta[c] = pd.to_numeric(mr_meta[c], errors=\"coerce\")\nmr_meta[\"contrast\"] = mr_meta[\"ContrastBolusAgent\"].fillna(\"\").str.strip() != \"\"\nmr_meta[\"Vendor\"] = mr_meta[\"Manufacturer\"].apply(vendor)\nmr_meta.to_csv(\"mr_meta.csv\", index=False)\n\n# Podsumowanie\nprint(\"Błędy odczytu:\", mr_meta.get(\"error\", pd.Series(dtype=str)).notna().sum(), \"\\n\")\nprint(\"Kontrast w nagłówku:\\n\", pd.crosstab(mr_meta[\"etykieta\"], mr_meta[\"contrast\"]), \"\\n\")\nprint(\"Płaszczyzna:\\n\", pd.crosstab(mr_meta[\"plaszczyzna\"].fillna(\"brak\"), mr_meta[\"etykieta\"]), \"\\n\")\nprint(\"2D/3D:\\n\", pd.crosstab(mr_meta[\"MRAcquisitionType\"].fillna(\"brak\"), mr_meta[\"etykieta\"]), \"\\n\")\nprint(\"Parametry (mediana i zakres):\")\nprint(mr_meta.groupby(\"etykieta\")[[\"RepetitionTime\", \"EchoTime\", \"InversionTime\", \"FlipAngle\", \"SliceThickness\"]]\n .describe().T.loc[(slice(None), [\"count\", \"min\", \"50%\", \"max\"]), :].round(1), \"\\n\")\nfor lab in [\"MRI T2\", \"MRI T1post\"]:\n print(f\"--- Najczęstsze opisy: {lab} ---\")\n print(mr_meta.loc[mr_meta[\"etykieta\"] == lab, \"SeriesDescription\"].str.upper().value_counts().head(20), \"\\n\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T13:04:02.935533Z","iopub.execute_input":"2026-09-24T13:04:02.935922Z","iopub.status.idle":"2026-09-24T13:05:05.143611Z","shell.execute_reply.started":"2026-09-24T13:04:02.935891Z","shell.execute_reply":"2026-09-24T13:05:05.142615Z"}},"outputs":[{"name":"stderr","text":" 0%| | 2/1288 [00:00<01:35, 13.52it/s]/usr/local/lib/python3.12/dist-packages/pydicom/valuerep.py:440: UserWarning: The value length (20) exceeds the maximum length of 16 allowed for VR SH.\n warn_and_log(msg)\n 1%| | 12/1288 [00:00<00:50, 25.35it/s]/usr/local/lib/python3.12/dist-packages/pydicom/valuerep.py:440: UserWarning: The value length (64) exceeds the maximum length of 16 allowed for VR SH.\n warn_and_log(msg)\n100%|██████████| 1288/1288 [01:02<00:00, 20.74it/s]","output_type":"stream"},{"name":"stdout","text":"Błędy odczytu: 0 \n\nKontrast w nagłówku:\n contrast False True \netykieta \nMRI T1post 108 197\nMRI T2 944 39 \n\nPłaszczyzna:\n etykieta MRI T1post MRI T2\nplaszczyzna \nczołowa 27 43\nosiowa 258 903\nstrzałkowa 20 37 \n\n2D/3D:\n etykieta MRI T1post MRI T2\nMRAcquisitionType \n2D 85 937\n3D 220 46 \n\nParametry (mediana i zakres):\netykieta MRI T1post MRI T2\nRepetitionTime count 305.0 818.0\n min 5.2 5.7\n 50% 500.0 4820.0\n max 3100.1 14535.6\nEchoTime count 305.0 818.0\n min 1.8 2.3\n 50% 3.2 100.0\n max 44.0 382.0\nInversionTime count 160.0 52.0\n min 0.0 0.0\n 50% 900.0 0.0\n max 1260.0 2500.0\nFlipAngle count 305.0 818.0\n min 8.0 15.0\n 50% 12.0 142.0\n max 160.0 180.0\nSliceThickness count 305.0 983.0\n min 0.8 0.5\n 50% 1.2 5.0\n max 5.0 6.0 \n\n--- Najczęstsze opisy: MRI T2 ---\nSeriesDescription\nT2 TSE 164\nAX T2 92\nAX T2 PROPELLER 68\nT2 AX FS 54\nAX T2 PROP 36\nT2 FS AX 35\nAX T2 FRFSE 32\nAX T2 TSE BLADE 31\nSAG T2 28\nCORONAL T2 24\nT2_TSE_TRA-FS 23\nT2_TSE_AX 20\nT2_TSE_FS_TRA 19\nAX T2 FSE 18\nT2W_TSE 16\nAX T2 +C 15\nT2 AX FS BLADE 15\nT2_TSE_TRA_FS 14\nAX T2 FS 12\nBRAIN AX T2_SE 10\nName: count, dtype: int64 \n\n--- Najczęstsze opisy: MRI T1post ---\nSeriesDescription\n3D AX T1 BRAVO+GD 20\nT1FS_3D_AX WHOLE BRAIN_GD. 20\nAX T1 +C 18\nGAD BRAIN AX 3D T1_GRE 13\nAX T1 MPR POST 8\nAX FSPGR 3D 8\n3D AX T1 BRAVO+GD FOR SRT 7\nVOLUMETRIC AXIAL GD 6\nST1W_3D_HR 6\nT1 FLAIR AX FS GD 5\nAX T1 FLAIR +C 4\nAX SPGR +C 4\nCOR T1 +C 4\nCOR OBL SPGR +C 4\nT1 FS AX GD 4\nVOLUMETRIC GD 4\n 0 else \"3D GRE (SPGR/VIBE)\"\n if d3 and ti > 0:\n return \"3D MPRAGE (Siemens)\"\n if d3:\n return \"3D TSE (SPACE/CUBE)\"\n if ti > 0:\n return \"2D T1 FLAIR\"\n return \"2D SE/TSE\"\n\ndef podtyp_t2(r):\n tr, ti = r[\"RepetitionTime\"], r[\"InversionTime\"] if pd.notna(r[\"InversionTime\"]) else 0\n if pd.isna(tr):\n return f\"brak TR (ImageType: {r['ImageType']})\"\n if ti > 1500:\n return \"FLAIR\"\n if tr < 100:\n return \"GRE (T2*/bSSFP)\"\n if r[\"MRAcquisitionType\"] == \"3D\":\n return \"3D T2 (SPACE/CUBE)\"\n return \"2D T2 TSE/FSE\"\n\nmr_meta[\"kontrast_dowod\"] = mr_meta.apply(dowod_kontrastu, axis=1)\nt1 = mr_meta[\"etykieta\"] == \"MRI T1post\"\nt2 = mr_meta[\"etykieta\"] == \"MRI T2\"\nmr_meta.loc[t1, \"podtyp\"] = mr_meta[t1].apply(podtyp_t1, axis=1)\nmr_meta.loc[t2, \"podtyp\"] = mr_meta[t2].apply(podtyp_t2, axis=1)\n\nprint(\"=== T1post: dowód kontrastu ===\")\nprint(mr_meta.loc[t1, \"kontrast_dowod\"].value_counts(), \"\\n\")\nprint(\"=== T1post: podtyp ===\")\nprint(mr_meta.loc[t1, \"podtyp\"].value_counts(), \"\\n\")\nprint(\"=== T1post bez żadnego dowodu kontrastu - opisy ===\")\nprint(mr_meta.loc[t1 & (mr_meta[\"kontrast_dowod\"] == \"brak dowodu\"), \"SeriesDescription\"].value_counts().head(20), \"\\n\")\n\nprint(\"=== T2: podtyp ===\")\nprint(mr_meta.loc[t2, \"podtyp\"].value_counts(), \"\\n\")\nprint(\"=== T2: kontrast ===\")\nprint(mr_meta.loc[t2, \"kontrast_dowod\"].value_counts(), \"\\n\")\nprint(\"=== T2 GRE i FLAIR - opisy ===\")\nprint(mr_meta.loc[t2 & mr_meta[\"podtyp\"].isin([\"GRE (T2*/bSSFP)\", \"FLAIR\"]),\n [\"podtyp\", \"SeriesDescription\", \"RepetitionTime\", \"EchoTime\", \"FlipAngle\"]].to_string(max_rows=40), \"\\n\")\n\n# Pole i producent\nmr_meta[\"Tesla\"] = pd.to_numeric(mr_meta[\"MagneticFieldStrength\"], errors=\"coerce\")\nmr_meta.loc[mr_meta[\"Tesla\"] > 100, \"Tesla\"] /= 10000\nprint(pd.crosstab(mr_meta[\"Tesla\"].round(1).fillna(-1), mr_meta[\"etykieta\"], margins=True), \"\\n\")\nprint(pd.crosstab(mr_meta[\"Vendor\"], mr_meta[\"etykieta\"], margins=True))\n\nmr_meta.to_csv(\"mr_meta_klasy.csv\", index=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-09-24T13:07:15.071608Z","iopub.execute_input":"2026-09-24T13:07:15.072591Z","iopub.status.idle":"2026-09-24T13:07:15.191655Z","shell.execute_reply.started":"2026-09-24T13:07:15.072555Z","shell.execute_reply":"2026-09-24T13:07:15.19081Z"}},"outputs":[{"name":"stdout","text":"=== T1post: dowód kontrastu ===\nkontrast_dowod\ntag w nagłówku 197\ntylko opis 77\nbrak dowodu 31\nName: count, dtype: int64 \n\n=== T1post: podtyp ===\npodtyp\n3D MPRAGE (Siemens) 73\n2D SE/TSE 67\n3D GRE (SPGR/VIBE) 66\n3D GRE z inwersją (BRAVO/MPRAGE-typ) 64\n2D T1 FLAIR 18\n3D TSE (SPACE/CUBE) 17\nName: count, dtype: int64 \n\n=== T1post bez żadnego dowodu kontrastu - opisy ===\nSeriesDescription\nsT1W_3D_HR 6\nAX CUBE T1 REFORMAT 1\nt1_space_sag_caipi4_iso 1\nT1 ax IR_FS 1\nAx T1 SE 1\n(C) Sag 3D T1 BRAVO 1\n3D T1 SAG 1\nName: count, dtype: int64 \n\n=== T2: podtyp ===\npodtyp\n2D T2 TSE/FSE 772\nbrak TR (ImageType: ORIGINAL\\PRIMARY\\T2\\NONE) 164\n3D T2 (SPACE/CUBE) 34\nGRE (T2*/bSSFP) 11\nFLAIR 1\nbrak TR (ImageType: ORIGINAL\\PRIMARY\\TOF\\NONE) 1\nName: count, dtype: int64 \n\n=== T2: kontrast ===\nkontrast_dowod\nbrak dowodu 939\ntag w nagłówku 39\ntylko opis 5\nName: count, dtype: int64 \n\n=== T2 GRE i FLAIR - opisy ===\n podtyp SeriesDescription RepetitionTime EchoTime FlipAngle\n121 GRE (T2*/bSSFP) t2_ci3d_tra_iso 6.1500 2.730 70.0\n166 GRE (T2*/bSSFP) ciss3d_tra_iso 8.2600 3.710 50.0\n174 GRE (T2*/bSSFP) AX CISS 6.8100 3.410 80.0\n328 GRE (T2*/bSSFP) t2_ci3d_tra_iso 6.0700 2.640 70.0\n377 GRE (T2*/bSSFP) ciss3d_tra_iso 5.7800 2.460 50.0\n486 GRE (T2*/bSSFP) Ax 3D FIESTA-C 5.7240 2.284 55.0\n520 FLAIR ax flair 9000.0000 137.000 150.0\n745 GRE (T2*/bSSFP) t2_ci3d_tra_iso 6.0700 2.640 70.0\n859 GRE (T2*/bSSFP) Ax 3D FIESTA-C 5.8640 2.360 55.0\n1032 GRE (T2*/bSSFP) Ax 3D FIESTA-C 6.7920 2.488 55.0\n1071 GRE (T2*/bSSFP) TRIGEM 150 slices 6.1714 3.086 60.0\n1218 GRE (T2*/bSSFP) None 7.3600 2.724 60.0 \n\netykieta MRI T1post MRI T2 All\nTesla \n-1.0 0 4 4\n 1.2 1 2 3\n 1.5 159 654 813\n 3.0 145 323 468\n All 305 983 1288 \n\netykieta MRI T1post MRI T2 All\nVendor \nCanon/Toshiba 0 16 16\nGE 124 295 419\nHitachi 1 2 3\nPhilips 28 254 282\nSiemens 152 416 568\nAll 305 983 1288\n","output_type":"stream"}],"execution_count":11},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}