|
a |
|
b/docs/tutorials/reason.md |
|
|
1 |
# Detecting Reason of Hospitalisation |
|
|
2 |
|
|
|
3 |
In this tutorial we will use the pipe `eds.reason` to : |
|
|
4 |
|
|
|
5 |
- Identify spans that corresponds to the reason of hospitalisation |
|
|
6 |
- Check if there are named entities overlapping with my span of 'reason of hospitalisation' |
|
|
7 |
- Check for all named entities if they are tagged `is_reason` |
|
|
8 |
|
|
|
9 |
```python |
|
|
10 |
import edsnlp, edsnlp.pipes as eds |
|
|
11 |
|
|
|
12 |
text = """COMPTE RENDU D'HOSPITALISATION du 11/07/2018 au 12/07/2018 |
|
|
13 |
MOTIF D'HOSPITALISATION |
|
|
14 |
Monsieur Dupont Jean Michel, de sexe masculin, âgée de 39 ans, née le 23/11/1978, |
|
|
15 |
a été hospitalisé du 11/08/2019 au 17/08/2019 pour attaque d'asthme. |
|
|
16 |
|
|
|
17 |
ANTÉCÉDENTS |
|
|
18 |
Antécédents médicaux : |
|
|
19 |
Premier épisode d'asthme en mai 2018.""" |
|
|
20 |
|
|
|
21 |
nlp = edsnlp.blank("eds") |
|
|
22 |
|
|
|
23 |
# Extraction d'entités nommées |
|
|
24 |
nlp.add_pipe( |
|
|
25 |
eds.matcher( |
|
|
26 |
terms=dict( |
|
|
27 |
respiratoire=[ |
|
|
28 |
"asthmatique", |
|
|
29 |
"asthme", |
|
|
30 |
"toux", |
|
|
31 |
] |
|
|
32 |
), |
|
|
33 |
), |
|
|
34 |
) |
|
|
35 |
nlp.add_pipe(eds.normalizer()) |
|
|
36 |
nlp.add_pipe(eds.sections()) |
|
|
37 |
nlp.add_pipe(eds.reason(use_sections=True)) |
|
|
38 |
|
|
|
39 |
doc = nlp(text) |
|
|
40 |
``` |
|
|
41 |
|
|
|
42 |
The pipe `reason` will add a key of spans called `reasons`. We check the first item in this list. |
|
|
43 |
|
|
|
44 |
```python |
|
|
45 |
# ↑ Omitted code above ↑ |
|
|
46 |
|
|
|
47 |
reason = doc.spans["reasons"][0] |
|
|
48 |
reason |
|
|
49 |
# Out: hospitalisé du 11/08/2019 au 17/08/2019 pour attaque d'asthme. |
|
|
50 |
``` |
|
|
51 |
|
|
|
52 |
Naturally, all spans included the `reasons` key have the attribute `#!python reason._.is_reason == True`. |
|
|
53 |
|
|
|
54 |
```python |
|
|
55 |
# ↑ Omitted code above ↑ |
|
|
56 |
|
|
|
57 |
reason._.is_reason |
|
|
58 |
# Out: True |
|
|
59 |
``` |
|
|
60 |
|
|
|
61 |
<!-- no-check --> |
|
|
62 |
|
|
|
63 |
```python |
|
|
64 |
# ↑ Omitted code above ↑ |
|
|
65 |
|
|
|
66 |
for e in reason._.ents_reason: # (1) |
|
|
67 |
print( |
|
|
68 |
"Entity:", |
|
|
69 |
e.text, |
|
|
70 |
"-- Label:", |
|
|
71 |
e.label_, |
|
|
72 |
"-- is_reason:", |
|
|
73 |
e._.is_reason, |
|
|
74 |
) |
|
|
75 |
# Out: Entity: asthme -- Label: respiratoire -- is_reason: True |
|
|
76 |
``` |
|
|
77 |
|
|
|
78 |
1. We check if the span include named entities, their labels and the attribute is_reason |
|
|
79 |
|
|
|
80 |
We can verify that named entities that do not overlap with the spans of reason, have their attribute `#!python reason._.is_reason == False`: |
|
|
81 |
|
|
|
82 |
```{ .python .no-check } |
|
|
83 |
for e in doc.ents: |
|
|
84 |
print(e.start, e, e._.is_reason) |
|
|
85 |
# Out: 42 asthme True |
|
|
86 |
# Out: 54 asthme False |
|
|
87 |
``` |