Repository navigation
Expand file tree
/
Copy pathindex.html
More file actions
318 lines (279 loc) · 14.3 KB
/
Copy pathindex.html
File metadata and controls
318 lines (279 loc) · 14.3 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
<!DOCTYPE html>
<html lang="ko">
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<head>
<meta charset="UTF-8">
<title>PU-Track</title>
<link rel="stylesheet" href="style.css">
</head>
<body>
<!-- Top Banner Header -->
<div style="background-color: #f5f5f5; width: 100%; padding: 30px 20px; text-align: center;">
<h1 style="color: #111; font-size: 2em; margin-bottom: 0.3em;">WE ARE PU-Track</h1>
<p style="color: #222; font-size: 1em; max-width: 800px; margin: 0 auto;">
A smart healthcare system that predicts optimal postural change intervals to prevent pressure ulcers<br>
using IoT sensing, machine learning, and large language models.
</p>
</div>
<!-- Info Section -->
<div class="section" style="display: flex; flex-wrap: wrap; gap: 40px; margin-top: 3em; justify-content: center;">
<!-- Affiliation Block -->
<div style="min-width: 280px; max-width: 400px;">
<h3 style="border-bottom: 2px solid #ccc; padding-bottom: 0.3em;">Affiliation</h3>
<p style="margin-top: 0.5em;">
Capstone Design II<br>
Department of Computer Science and Engineering,<br>
Sogang University
</p>
</div>
<!-- Recognitions Block -->
<div style="min-width: 280px; max-width: 500px;">
<h3 style="border-bottom: 2px solid #ccc; padding-bottom: 0.3em;">Recognitions</h3>
<ul style="margin-top: 0.5em; padding-left: 1.2em; line-height: 1.8;">
<li>Paper accepted at <strong>2025 Korea Computer Congress (KCC)</strong> – IoT Track</li>
<li>4th Prize at <strong>2025-1 Sogang Capstone Design Competition</strong></li>
<li>Finalist at <strong>K-Hackton 2025</strong> (ongoing)</li>
</ul>
</div>
</div>
<hr style="margin-top: 3em;">
<!-- 프로젝트 이유 -->
<h2>0. Motivation</h2>
<div class="section">
<p>
We were deeply struck by the fact that pressure ulcers — a condition experienced by many immobile patients — can
lead to a mortality rate as high as 70%, despite being preventable.
Prevention largely depends on two key actions: <strong>regular postural changes and maintaining an appropriate
environment</strong>.
However, doing so manually places a <strong>heavy burden on caregivers</strong> and medical staff.
Motivated by this, we set out to develop an <strong>automated pressure ulcer prevention system</strong> that can
monitor and manage
risk factors without constant human intervention.
</p>
</div>
<hr>
<!-- 프로젝트 설명 -->
<h2>1. Overview (Detailed)</h2>
<div class="section">
<h3>👤 User Flow</h3>
<p>
<ol>
<li> The user signs up through the web portal and receives a unique patient code.</li>
<li> Using the issued code, they log in on the mobile app.</li>
<li> Caregivers can access three core features:</li>
</ol>
</p>
<br>
<h3>📌 Main Features</h3>
<li>
<strong>1️⃣ Real-Time Dashboard & Alerts (IoT)</strong><br>
Monitors patient info and sensor data (temp, humidity) in real time.<br>
Sends alerts for abnormal readings or inactivity.
</li>
<li>
<strong>2️⃣ Posture Change Prediction (ML)</strong><br>
Predicts optimal timing for the next posture shift based on recent sensor trends.
</li>
<li>
<strong>3️⃣ Daily Summary Report (LLM)</strong><br>
At sleep mode, generates a Korean summary using 24h logs + weekly trends,<br>
including personalized recommendations for patient care.
</li>
<br>
<h3>🔍 Technical Details</h3>
<div style="text-align: center; margin-bottom: 20px;">
<img src="./attachment/service_architecture.png" alt="System Architecture"
style="max-width: 100%; height: auto; border: 0px solid #ccc; border-radius: 8px;">
<p style="font-size: 0.95em; color: #555;">▲ Overall System Architecture</p>
</div>
<br>
<!-- IoT 아키텍처 및 센서 설명 이미지 -->
<h4>1️⃣ IoT Logic (ESP32 + MQTT)</h4>
<div style="display: flex; justify-content: center; gap: 20px; flex-wrap: wrap; margin: 15px 0;">
<img src="./attachment/iot_architecture.png" alt="IoT Architecture"
style="max-width: 45%; min-height: 250px; height: auto; object-fit: contain; border: 1px solid #ccc; border-radius: 6px;">
<img src="./attachment/iot_sensor.png" alt="Sensor Diagram"
style="max-width: 45%; min-height: 250px; height: auto; object-fit: contain; border: 0px solid #ccc; border-radius: 6px;">
</div>
<ul style="padding-left: 1.5em; list-style-type: disc; font-size: 1em; line-height: 1.8;">
<li>The ESP32 module collects data every 5 minutes from the DS18B20, DHT22, and RFID-RC522 sensors.</li>
<li>These sensor values are formatted as JSON and transmitted to <strong>both the MQTT broker (HiveMQ) and
Firebase Realtime Database</strong>.</li>
<li>The MQTT broker <strong>instantly forwards the real-time data to the iOS app</strong>, enabling live
monitoring.</li>
<li>The data stored in Firebase is later sent to the backend server for logging and further analysis.</li>
<br>
<li style="list-style-type: none; margin-top: 10px;">
<div style="border: 1px solid #ccc; background-color: #f0f4ff; padding: 12px 15px; border-radius: 10px;">
<strong>Why choose ESP32 over Arduino?</strong><br>
ESP32 was chosen over a traditional Arduino due to its built-in Wi-Fi capabilities, lower power consumption,
and suitability for wireless IoT communication in long-term deployments.
</div>
</li>
</ul>
<br>
<h4>2️⃣ Machine Learning (Random Forest)</h4>
<div style="display: flex; justify-content: center; gap: 20px; flex-wrap: wrap; margin: 15px 0;">
<img src="./attachment/MLcode.png" alt="ML code"
style="max-width: 45%; min-height: 250px; height: auto; object-fit: contain; border: 0px solid #ccc; border-radius: 6px;">
<p style="text-align: center; font-size: 1.1em; margin: 1.5em 0;">
\[
\frac{1}{M} \sum_{i=t-M}^{t} |\Delta T_i| < \varepsilon, \quad \text{with } M=5,\ \varepsilon=0.1^\circ C \]
</p>
</div>
<ul style="padding-left: 1.5em; list-style-type: disc; font-size: 1em; line-height: 1.8;">
<li>A Random Forest model was trained using sensor-derived features, including recent temperature stability
patterns, to predict the optimal timing for the next posture change.</li>
<li>The concept of "near-high-temperature stability" was used to identify sustained pressure risk conditions
based
on minimal temperature variation across recent intervals.</li>
<li>Model inference is executed in the backend (FastAPI) whenever a new posture is tagged.</li>
<br>
<li style="list-style-type: none; margin-top: 10px;">
<div style="border: 1px solid #ccc; background-color: #f0f4ff; padding: 12px 15px; border-radius: 10px;">
<strong>Why Random Forest?</strong><br>
<div style="display: flex; justify-content: center; gap: 20px; flex-wrap: wrap; margin: 15px 0;">
<img src="./attachment/ML_performance.png" alt="IoT Architecture"
style="max-width: 45%; min-height: 250px; height: auto; object-fit: contain; border: 0px solid #ccc; border-radius: 6px;">
<img src="./attachment/ML_graph.png" alt="Sensor Diagram"
style="max-width: 45%; min-height: 250px; height: auto; object-fit: contain; border: 0px solid #ccc; border-radius: 6px;">
</div>
To provide personalized predictions for a diverse patient population,
we explored alternative models such as LightGBM.
However, it exhibited inferior performance in our evaluation compared to Random Forest.
Random Forest was ultimately selected due to its strong performance on small-to-medium-sized datasets,
robustness to noise, and ability to model non-linear interactions across multiple sensor inputs
while minimizing overfitting through its ensemble structure.
</div>
</li>
</ul>
<br>
<h4>3️⃣ Daily Summary via LLM</h4>
<div style="display: flex; justify-content: center; gap: 20px; flex-wrap: wrap; margin: 15px 0;">
<img src="./attachment/LLMcode.png" alt="LLM code"
style="max-width: 70%; min-height: 250px; height: auto; object-fit: contain; border: 0px solid #ccc; border-radius: 6px;">
</div>
<ul style="padding-left: 1.5em; list-style-type: disc; font-size: 1em; line-height: 1.8;">
<li>When a patient transitions to sleep mode, the backend collects sensor logs from the past 24 hours and
constructs a Korean-language prompt for GPT-4 generation.</li>
<li>The prompt includes not only today's air and cushion temperature patterns, posture changes, and patient
metadata, but also comparative trends from the previous week.</li>
<li>GPT-4 generates a daily summary in fluent Korean, integrating medical insights, behavioral patterns, and
suggested actions to improve the patient's condition.</li>
<li>The result is displayed in natural language within the iOS caregiver app, helping caregivers make informed
decisions without interpreting raw sensor data.</li>
<li>Each summary includes personalized guidance such as recommended posture change timing or environmental
adjustments (e.g., cushion temperature), enhancing proactive care.</li>
<li>Summaries are generated automatically at sleep transition and optionally stored in Firebase for audit trails
or integration with electronic medical records (EMRs).</li>
<br>
<li style="list-style-type: none; margin-top: 10px;">
<div style="border: 1px solid #ccc; background-color: #f0f4ff; padding: 12px 15px; border-radius: 10px;">
<strong>Why use an LLM?</strong><br>
Even for trained medical professionals,
interpreting raw numerical logs alone can be challenging
when assessing a patient’s condition.
By leveraging a Large Language Model (LLM),
we transform structured sensor data into intuitive,
human-readable summaries that enhance situational awareness and clinical decision-making.
</div>
</li>
</ul>
<hr>
<h2>2. Demo Video</h2>
<div class="section">
<video controls>
<source src="./pdf/demoVideo.mp4" type="video/mp4">
현재 브라우저는 비디오 태그를 지원하지 않습니다.
</video>
</div>
<hr>
<!-- Tech Stack Section -->
<h2><strong>3. Link</strong></h2>
<div class="section">
<ul>
<li><a href="./pdf/first.pdf" target="_blank">📢 First Presentation</a></li>
<li><a href="./pdf/middle.pdf" target="_blank">📢 Middle Presentation</a></li>
<li><a href="./pdf/final.pdf" target="_blank">📢 Final Presentation</a></li>
<li><a href="./pdf/sogangcontest.pdf" target="_blank">🎉 Sogang Contest Poster</a></li>
<li><a href="./pdf/thesis.pdf" target="_blank">📄 KCC Thesis</a></li>
</ul>
</div>
<hr>
<div class="section">
<ul>
<li>🔗 <a href="https://github.com/PU-Track" target="_blank"><strong>GitHub Organization</strong></a></li><br>
<ul>
<li>🖇️ <a href="https://github.com/PU-Track/client" target="_blank">Web (User Registration)</a>
<br>
<span class="tech-badge">HTML</span>
<span class="tech-badge">CSS</span>
<span class="tech-badge">JavaScript</span>
<span class="tech-badge">Vite</span>
</li>
<br>
<li>🖇️ <a href="https://github.com/PU-Track/iOS" target="_blank">Mobile (Service App)</a>
<br>
<span class="tech-badge">Swift</span>
<span class="tech-badge">SwiftUI</span>
<span class="tech-badge">MQTT</span>
<span class="tech-badge">Firebase</span>
</li>
<br>
<li>🖇️ <a href="https://github.com/PU-Track/iot" target="_blank">IoT (ESP32 Firmware)</a>
<br>
<span class="tech-badge">C++</span>
<span class="tech-badge">ESP32</span>
<span class="tech-badge">MQTT</span>
</li>
<br>
<li>🖇️<a href="https://github.com/PU-Track/server" target="_blank">Backend (API Server)</a>
<br>
<span class="tech-badge">JavaSpring</span>
<span class="tech-badge">Docker</span>
<span class="tech-badge">FastAPI</span>
<span class="tech-badge">Firebase Admin SDK</span>
</li>
<br>
<li>🖇️ <a href="https://github.com/PU-Track/model" target="_blank">Model (ML Inference)</a>
<br>
<span class="tech-badge">Python</span>
<span class="tech-badge">Docker</span>
<span class="tech-badge">scikit-learn</span>
<span class="tech-badge">Random Forest</span>
</li>
<br>
<li>🖇️ <a href="https://github.com/PU-Track/docs" target="_blank">Docs (Documentation)</a>
<br>
<span class="tech-badge">Markdown</span>
<span class="tech-badge">HTML</span>
<span class="tech-badge">CSS</span>
</li>
<br>
</ul>
</li>
</ul>
</div>
<hr>
<h2><strong>4. Team</strong></h2>
<div class="member-container">
<div class="member">
<img src="./attachment/jiwon.png" alt="Jiwon Shin" class="member-photo"><br>
<strong>Jiwon Shin</strong> (Leader) - IoT & Front<br>
<a href="https://github.com/ena-isme" target="_blank"> @ena-isme</a><br>
<span class="function-badge">Web</span>
<span class="function-badge">Mobile</span>
<span class="function-badge">IoT</span>
<span class="function-badge">Docs</span>
</div>
<div class="member">
<img src="./attachment/jeongyeon.png" alt="Jeongyeon Lim" class="member-photo"><br>
<strong>Jeongyeon Lim</strong> - Back<br>
<a href="https://github.com/13155a1" target="_blank"> @13155a1</a><br>
<span class="function-badge">Server</span>
<span class="function-badge">ML</span>
</div>
</div>
</body>
</html>