An Embedding-Based Approach for Oral Disease Diagnosis Prediction - - PowerPoint PPT Presentation

▶
an embedding based approach for oral disease diagnosis
SMART_READER_LITE
LIVE PREVIEW

An Embedding-Based Approach for Oral Disease Diagnosis Prediction - - PowerPoint PPT Presentation

Oct 29, 2023 287 likes •544 views

An Embedding-Based Approach for Oral Disease Diagnosis Prediction from Electronic Medical Records Source: ICMHI 2018 Advisor: Jia-Ling Koh Speaker: Chih-Hsuan Tzang Date: 2019/3/18 1 Introduction Method Experiment

slide-1
SLIDE 1

An Embedding-Based Approach for Oral Disease Diagnosis Prediction from Electronic Medical Records

  • 1

Source: ICMHI 2018 Advisor: Jia-Ling Koh Speaker: Chih-Hsuan Tzang Date: 2019/3/18

slide-2
SLIDE 2
  • Introduction
  • Method
  • Experiment
  • Conclusion

2

slide-3
SLIDE 3
  • Introduction
  • Method
  • Experiment
  • Conclusion

3

slide-4
SLIDE 4

Introduction

4

Motivation

  • EHRs have been a valuable medical asset that can be used to

facilitate clinical decision making by discovering hidden knowledge and predicting diagnoses

  • Often fail to Capture the correlations inherent in patient clinical

data.

slide-5
SLIDE 5

Introduction

5

Goal

  • we intend to exploit the PKUSS data by exploring state-of-the-art

AI technologies that can capture correlations hidden in the EMRs

slide-6
SLIDE 6
  • Introduction
  • Method
  • Experiment
  • Conclusion

6

slide-7
SLIDE 7

Method

7

Data Description:

  • 7208 PKUSS EMRs
  • EMRs belong to the 11 departments
  • Each record consist 6 parts.

Departments Pediatric Dentistry Oral and Maxillofacial Surgery Laser Dentistry Emergency Oral Medicine Prosthodontics Geriatric Dentistry General Dentistry Orthodontics Periodontology Implant Dentistry Parts Chief complaints (C.C) family history (F .H.) History of present illness (HPI) Physical examination (P .E.) Past medical history (PMH) Diagnosis

slide-8
SLIDE 8

Method

8

Feature extraction

  • Extract symptoms based on

bigram from medical history C.C., HPI, PMH, and F .H.

  • Regular expression to

match symptoms from P .E. Learning Symptom and Diagnosis

slide-9
SLIDE 9

Method

9

Feature Extraction

slide-10
SLIDE 10

Method

10

Learning the correlation

slide-11
SLIDE 11

Method

11

A EMR dataset:

  • Feature vectors
  • Diagnosis

{(xi, yi)}m

i=1

xi yi

xi = (x(1)

i , . . . , x(n) i )

yi ∈ {1,...,l}

Sa Sb Sc Sd

  • ex. = (0, 1, 0 , 1)

x1

Learning the correlation

slide-12
SLIDE 12

Method

12

xi ⊙ C = (x(1)

i C(symp1), . . . , x(n) i C(sympn))𝖴

C( ⋅ ) ∈ ℝk

k k k k

Sa Sb Sc Sd

  • ex. = (0, 1, 0 , 1)

x1

k k k k

xi ⊙ C

(n × k)

Learning the correlation

slide-13
SLIDE 13

Method

13

a1 = tanh(z1)

W1

k k k k

z1 = W1(xi ⊙ C) + B1

(n × k) × 1

k × 1

k × (n × k)

k × 1

k × 1

Learning the correlation

slide-14
SLIDE 14

Method

14

yi ⊙ V = (y(1)

i V(diag1), . . . , y(l) i V(diagl))𝖴

W2 D1 D2 D3 D4 D5

yi ∈ {1,...,l}

k k k k k

V( ⋅ )

(l × k)

z2 = W2a1 + B2

l × k

k × 1

l × 1 l × 1

Learning the correlation

slide-15
SLIDE 15

Method

15

Learning the correlation

Softmax:

  • ij =

e(xiC)V[j]T−Uij ∑l

k=1 e(xiC)V[k]T−Uik

Uij =∥ xiC ∥ xi ∥1 − V[j] ∥2

slide-16
SLIDE 16
  • Introduction
  • Method
  • Experiment
  • Conclusion

16

slide-17
SLIDE 17

17

Experiment

ACC = t Nsum × 100

slide-18
SLIDE 18

18

Experiment

Weighted_P =

l

∑

i=1

αiPi Weighted_R =

l

∑

i=1

αiRi

Weighted_F1 =

l

∑

i=1

αiF1i

Precision = TP TP + FP

Recall = TP TP + FN

F1 = 2(P * R) P + R

slide-19
SLIDE 19

19

Experiment

K = p0 − pe 1 − pe

slide-20
SLIDE 20

20

Experiment

idfi = log |D| 1 + |{j : sympi ∈ diagj}| tfi,j = ni,j ∑k nk,j

tfidfi,j = tfi,j × idfi

slide-21
SLIDE 21

21

Experiment

slide-22
SLIDE 22

22

Experiment

slide-23
SLIDE 23
  • Introduction
  • Method
  • Experiment
  • Conclusion

23

slide-24
SLIDE 24

Conclusion

24

  • Build the model by learning the both symptoms and

diagnosis embeddings to predict the diagnosis.

  • Expressing the unique correlations that exist in the medical

records.