app.py

import streamlit as st
import pandas as pd
import numpy as np
from math import ceil
from collections import Counter
from string import punctuation
import spacy
from spacy import displacy
from spacy.lang.en import English
from spacy.matcher import PhraseMatcher
from spacy.tokens import Span
from negspacy.negation import Negex
#import en_ner_bc5cdr_md
import re


from streamlit.components.v1 import html


# Store the initial value of widgets in session state
if "visibility" not in st.session_state:
    st.session_state.visibility = "visible"
    st.session_state.disabled = False

#nlp = en_core_web_lg.load()
#nlp = spacy.load("en_ner_bc5cdr_md")
nlp0 = spacy.load("en_core_sci_sm")
nlp1 = spacy.load("en_ner_bc5cdr_md")

st.set_page_config(page_title ='Clinical Note Summarization', 
                   #page_icon= "Notes",
                   layout='wide')
st.title('Clinical Note Summarization')
st.markdown(
    """
    <style>
    [data-testid="stSidebar"][aria-expanded="true"] > div:first-child {
        width: 400px;
    }
    [data-testid="stSidebar"][aria-expanded="false"] > div:first-child {
        width: 400px;
        margin-left: -230px;
    }
    </style>
    """,
    unsafe_allow_html=True,
)
st.sidebar.markdown('Using transformer model')

## Loading in dataset
#df = pd.read_csv('mtsamples_small.csv',index_col=0)
df = pd.read_csv('shpi_w_rouge21Nov.csv')
#Loading in Admission chief Complaint and diagnosis
df2 = pd.read_csv('cohort_cc_adm_diag.csv')

#combining both data into one 
df = pd.merge(df, df2, on=['HADM_ID','SUBJECT_ID'])

df['HADM_ID'] = df['HADM_ID'].astype(str).apply(lambda x: x.replace('.0',''))


#Renaming column
df.rename(columns={'SUBJECT_ID':'Patient_ID',
                  'HADM_ID':'Admission_ID',
                  'hpi_input_text':'Original_Text',
                  'hpi_reference_summary':'Reference_text'}, inplace = True)
 
 #data.rename(columns={'gdp':'log(gdp)'}, inplace=True)

#Filter selection 
st.sidebar.header("Search for Patient:")

patientid = df['Patient_ID']
patient = st.sidebar.selectbox('Select Patient ID:', patientid)
admissionid = df['Admission_ID'].loc[df['Patient_ID'] == patient]
HospitalAdmission = st.sidebar.selectbox(' ', admissionid) 

# List of Model available
model = st.sidebar.selectbox('Select Model', ('BertSummarizer','BertGPT2','t5seq2eq','t5','gensim','pysummarizer'))

col3,col4 = st.columns(2) 
patientid = col3.write(f"Patient ID:  {patient} ")
admissionid =col4.write(f"Admission ID:  {HospitalAdmission} ")
runtext = ''
inputNote  ='Input note here:'
# Query out relevant Clinical notes
original_text =  df.query(
    "Patient_ID  == @patient & Admission_ID == @HospitalAdmission"
)
original_text2 = original_text['Original_Text'].values
AdmissionChiefCom = original_text['Admission_Chief_Complaint'].values
diagnosis =original_text['DIAGNOSIS'].values
reference_text = original_text['Reference_text'].values


##========= Buttons to the 4 tabs ========
col1, col2, col3, col4, col5 = st.columns([1,1,1,1,1])
col6, col7, col8 =st.columns([2,2,2])
with st.container():
    with col1:
        btnAdmission = st.button("🏥 Admission")
        if btnAdmission:
            #nav_page('Admission')
            inputNote = "Input Admission Note"
    with col2:
        btnDailyNarrative = st.button('📆Daily Narrative')
        if btnDailyNarrative:
            inputNote = "Input Daily Narrative Note"
    with col3:
        btnDischargePlan = st.button('🗒️Discharge Plan')
        if btnDischargePlan:
            inputNote = "Input Discharge Plan"
    with col4:
        btnSocialNotes = st.button('📝Social Notes')
        if btnSocialNotes: 
            inputNote = "Input Social Note"
    with col5:
        btnPastHistory = st.button('📇Past History (6 Mths)')
        if btnPastHistory:          
            inputNote = "Input History records"
            
if btnPastHistory:
    st.text_area('Past History','Date of discharge: xxxxxxxxx')
    
else:
    runtext =st.text_area(inputNote, str(original_text2), height=300)
    
    
with st.container(): 
    if btnPastHistory:
        with col6: 
            st.markdown('**No. of admission past 6 months: xx**')
        with col7: 
            st.text_area('Discharge Disposition:',' ', height=8) #to replace with dropdown list if data is available 
        with col8:
            st.date_input('Select Admission Date') # To replace with a dropdown filter instead 

    



# Extract words associated with each entity
def genEntities(ann, entity):
              # entity colour dict
              #ent_col = {'DISEASE':'#B42D1B', 'CHEMICAL':'#F06292'}
              ent_col = {'DISEASE':'pink', 'CHEMICAL':'orange'}
              # separate into the different entities
              entities = trans_df['Class'].unique()

              if entity in entities:
                             ent = list(trans_df[trans_df['Class']==entity]['Entity'].unique())
                             entlist = ",".join(ent)
                             st.markdown(f'<p style="background-color:{ent_col[entity]};color:#080808;font-size:16px;">{entlist}</p>', unsafe_allow_html=True)
                                                                       
#lemmatizing the notes to capture all forms of negation(e.g., deny: denies, denying)
def lemmatize(note, nlp):
    doc = nlp(note)
    lemNote = [wd.lemma_ for wd in doc]
    return " ".join(lemNote)

#function to modify options for displacy NER visualization
def get_entity_options():
    entities = ["DISEASE", "CHEMICAL", "NEG_ENTITY"]
    colors = {'DISEASE': 'pink', 'CHEMICAL': 'orange', "NEG_ENTITY":'white'}
    options = {"ents": entities, "colors": colors}    
    return options

#adding a new pipeline component to identify negation
def neg_model(nlp_model):
    nlp = spacy.load(nlp_model, disable = ['parser'])
#     nlp.add_pipe(nlp.create_pipe('sentencizer'))
    nlp.add_pipe('sentencizer')
#     negex = Negex(nlp)
    nlp.add_pipe(
    "negex",
    config={
        "chunk_prefix": ["no"],
    },
    last=True)
    return nlp
"""
Negspacy sets a new attribute e._.negex to True if a negative concept is encountered
"""
def negation_handling(nlp_model, note, neg_model):
    results = []
    nlp = neg_model(nlp_model) 
    note = note.split(".") #sentence tokenizing based on delimeter 
    note = [n.strip() for n in note] #removing extra spaces at the begining and end of sentence
    for t in note:
        doc = nlp(t)
        for e in doc.ents:
            rs = str(e._.negex)
            if rs == "True": 
                results.append(e.text)
    return results

#function to identify span objects of matched negative phrases from text
def match(nlp,terms,label):
    patterns = [nlp.make_doc(text) for text in terms]
    matcher = PhraseMatcher(nlp.vocab)
    matcher.add(label, None, *patterns)
    return matcher

#replacing the labels for identified negative entities
def overwrite_ent_lbl(matcher, doc):
    matches = matcher(doc)
    seen_tokens = set()
    new_entities = []
    entities = doc.ents
    for match_id, start, end in matches:
        if start not in seen_tokens and end - 1 not in seen_tokens:
            new_entities.append(Span(doc, start, end, label=match_id))
            entities = [e for e in entities if not (e.start < end and e.end > start)]
            seen_tokens.update(range(start, end))
    doc.ents = tuple(entities) + tuple(new_entities)
    return doc

#deduplicate repeated entities 
def dedupe(items):
    seen = set()
    for item in items:
        item = str(item).strip()
        if item not in seen:
            yield item
            seen.add(item)

lem_clinical_note= lemmatize(text, nlp0)
#creating a doc object using BC5CDR model
doc = nlp1(lem_clinical_note)
options = get_entity_options()

#list of negative concepts from clinical note identified by negspacy
results0 = negation_handling("en_ner_bc5cdr_md", lem_clinical_note, neg_model)

matcher = match(nlp1, results0,"NEG_ENTITY")

#doc0: new doc object with added "NEG_ENTITY label"
doc0 = overwrite_ent_lbl(matcher,doc)

#visualizing identified Named Entities in clinical input text 
displacy.render(doc0, style='ent', options=options)

##=== end of NER tagger===
def visualize (run_text,output):
    text =''
    splitruntext = [x for x in runtext.split('.')]
    splitoutput = [x for x in output.split('.')]
    return splitoutput,splitruntext
    

def run_model(input_text):    
    if model == "BertSummarizer":
        output = original_text['BertSummarizer'].values
        st.write('Summary')

    elif model == "BertGPT2":
        output = original_text['BertGPT2'].values
        st.write('Summary')
        
      
    elif model == "t5seq2eq": 
        output = original_text['t5seq2eq'].values
        st.write('Summary')
         
    elif model == "t5":
        output = original_text['t5'].values
        st.write('Summary')
        
    elif model == "gensim": 
        output = original_text['gensim'].values
        st.write('Summary')
        
    elif model == "pysummarizer": 
        output = original_text['pysummarizer'].values
        st.write('Summary')


    
    st.success(output)

doc = nlp(str(original_text2))
colors = { "DISEASE": "pink","CHEMICAL": "orange"}
options = {"ents": [ "DISEASE", "CHEMICAL"],"colors": colors}
ent_html = displacy.render(doc, style="ent", options=options)

col1, col2 = st.columns([1,1])

with col1:
    if not(btnPastHistory): #to not show summary and references text for Past History
        st.button('Summarize')
        run_model(runtext)
        #sentences=runtext.split('.')
        st.text_area('Reference text', str(reference_text), height=150)
    else: 
        with st.expander('Full Discharge Summary'):
            st.write( str(original_text2))
    ##====== Storing the Diseases/Text
    table= {"Entity":[], "Class":[]}
    ent_bc = {}
    for x in doc.ents:
        ent_bc[x.text] = x.label_
    for key in ent_bc:
        table["Entity"].append(key)
        table["Class"].append(ent_bc[key])
    trans_df = pd.DataFrame(table)

with col2:
    st.button('NER')
    st.markdown('**CHIEF COMPLAINT:**')
    st.write(str(AdmissionChiefCom))
    st.markdown('**ADMISSION DIAGNOSIS:**')
    st.markdown(str(diagnosis))
    st.markdown('**PROBLEM/ISSUE**')
    genEntities(trans_df, 'DISEASE')
    st.markdown('**MEDICATION**')
    genEntities(trans_df, 'CHEMICAL')
    #st.table(trans_df)   
    st.markdown('**NER**')
    with st.expander("See NER Details"):
        st.markdown(ent_html, unsafe_allow_html=True)