Predicting Respiratory Disease from Age, Gender, Symptom

Creating and deploying a model as described in https://github.com/alexeygrigorev/mlbookcamp-code/

Disclaimer: The only purpose of this code repository is creating and deploying an example statistical model. No medical advice. The underlying data my be wrong and any conclusion made false.

Dataset: https://www.kaggle.com/datasets/abbotpatcher/respiratory-symptoms-and-treatment

• 1st Notebook for EDA and initial Model: https://www.kaggle.com/code/bnzn261029/midterm-resp
• 2nd Notebook with cleaned data (i.e. symptom list) and rebuild, second Model: https://www.kaggle.com/code/bnzn261029/fork-of-midterm-resp-symptoms-cleaned
• 3rd Notebook with Decision Tree: https://www.kaggle.com/code/bnzn261029/midterm-resp-decision-tree

Problem Description

In medicine finding the right diagnosis is key for successful treatment and thus restituation of health. Diagnosing a disease involves several steps. It includes history taking, physical examination and diagnostic procedures followed by a preliminary list of possible diagnoses. Then estimation of the most likely diagnosis can be made through different statistic techniques and experienced clinicians develop an instinct to come to a conclusion fast. But improving the diagnostic process and reducing uncertainty is an ongoing challenge and offers opportunities for data science solutions.

Ahsan MM, Luna SA, Siddique Z. Machine-Learning-Based Disease Diagnosis: A Comprehensive Review. Healthcare (Basel). 2022 Mar 15;10(3):541. doi: 10.3390/healthcare10030541. PMID: 35327018; PMCID: PMC8950225.

Raita Y, Camargo CA Jr, Liang L, Hasegawa K. Big Data, Data Science, and Causal Inference: A Primer for Clinicians. Front Med (Lausanne). 2021 Jul 6;8:678047. doi: 10.3389/fmed.2021.678047. PMID: 34295910; PMCID: PMC8290071.

Example

Input (as used in test_service.py)

patient = {
	'Age': 60,
	'Sex=female': 0, 
	'Sex=male': 1, 
	'Sex=not to say': 0,
	'Symptoms_encoded': 48,
}

Or send curl request to Flask API https://bsenst.pythonanywhere.com/predict (availability not guaranteed)

curl --location --request POST 'https://bsenst.pythonanywhere.com/predict' \
--header 'Content-Type: application/json' \
--data-raw '{"Age": 60, "Sex=female": 0, "Sex=male": 1, "Sex=not to say": 0, "Symptoms_encoded": 48}'

Output (response from the model served as flask app):

{
  "disclaimer": "This script is for educational purpose only.",
  "disease": "Pneumonia",
  "disease_probability": 0.19968054490140236,
  "patient:": {
	"Age": 60,
	"Sex=female": 0,
	"Sex=male": 1,
	"Sex=not to say": 0,
	"Symptoms_encoded": 48
  }
}

This version of the model suffers from low accuracy. For a male 60 years old patient with a cold it predicts pneumonia with a probability of 20 %. If the age is changed to 10 years old it suggests bronchitis with a probability of 22 %.

Output:

{
  "disclaimer": "This script is for educational purpose only.",
  "disease": "bronchitis",
  "disease_probability": 0.2179886636147549,
  "patient:": {
	"Age": 10,
	"Sex=female": 0,
	"Sex=male": 1,
	"Sex=not to say": 0,
	"Symptoms_encoded": 48
  }
}

Run the Service using Python

Create virtual environment (as described in https://docs.python.org/3/tutorial/venv.html)

python3 -m venv virtualenv

Then activate the virtual environment for Windows

virtualenv\Scripts\activate.bat

Or on Unix/MacOS

source virtualenv/bin/activate

Install requirements

pip install -r requirements.txt

Run the flask app

python flask_app.py

Open another command window and run the example post request

python test_service.py

Make sure the test script sends the request to the url the flask app is being served (i.e. https://172.17.0.2:9696).

Run the Service using Docker

Download the docker image (206.6 MB)

docker pull fritz.jfrog.io/default-docker-local/midterm-resp-docker:latest

Compare the checksum sha256 78583fdc866a728a2d3588d3d7c45b44b7a9f9a9c2175e81688d9ec5ab6d5a42

Build and run the docker image

docker build --tag midterm-resp-docker

docker run midterm-resp-docker

You should see that the flask app is being served. Open another terminal and run the test script.

python test_service.py

Make sure the test script sends the request to the url the flask app is being served (i.e. https://172.17.0.2:9696).

Labels

LabelEncoder() Disease:

Label	Disease
0	Acute Respiratory Distress Syndrome
1	Asbestosis
2	Aspergillosis
3	Asthma
4	Bronchiectasis
5	Chronic Bronchitis
6	Influenza
7	Mesothelioma
8	Pneumonia
9	Pneumothorax
10	Pulmonary hypertension
11	Respiratory syncytial virus
12	Tuberculosis
13	bronchiolitis
14	bronchitis
15	chronic obstructive pulmonary disease
16	sleep apnea

LabelEncoder() Symptoms:

Label	Symptom
0	coughing
1	coughing
2	fatigue
3	low energy
4	shortness of breath
5	wheezing
6	A cough that lasts more than three weeks
7	A dry, crackling sound in the lungs while breathing in
8	Bluish skin
9	Chest congestion
10	Chest pain
11	Chills
12	Coughing up blood
13	Coughing up yellow or green mucus daily
14	Daytime sleepiness
15	Difficulties with memory and concentration
16	Dry mouth
17	Fatigue
18	Fatigue, feeling run-down or tired
19	Feeling run-down or tired
20	Fever
21	Frequently waking
22	Headache
23	Loss of appetite
24	Loss of appetite and unintentional weight loss
25	Low-grade fever
26	Morning headaches
27	Nasal congestion
28	Nausea
29	Night sweats
30	Pauses in breathing
31	Persistent dry coug
32	Persistent dry cough
33	Rapid breathing
34	Rapid heartbeat
35	Runny nose
36	Shortness of breath
37	Shortness of breath that gets worse during flare-ups
38	Snoring
39	Sore throat
40	Unusual moodiness
41	Weight loss from loss of appetite
42	Wheezing
43	Wider and rounder than normal fingertips and toes
44	allergy
45	breath
46	chest pain
47	chronic cough
48	cold
49	cough
50	cough with blood
51	coughing
52	diarrhea
53	distressing
54	dizziness
55	dry cough
56	edema
57	fainting
58	faster heart beating
59	fatigue
60	fever
61	greenish cough
62	heart palpitations
63	high fever
64	irritability
65	joint pain
66	loss of appetite
67	low energy
68	lower back pain
69	mucus
70	muscle aches
71	nausea
72	pain
73	runny nose
74	shaking
75	shallow breathing
76	sharp chest pain
77	short of breath
78	short, shallow and rapid breathing
79	shortness of breath
80	stuffy nose
81	sweating
82	tight feeling in the chest
83	vomiting
84	weight loss
85	wheezing
86	wheezing cough
87	whistling sound while breathing
88	whistling sound while you breathe
89	yellow cough