Predicting Disease Progression With Artificial Intelligence Assignment Sample

• Type Assignment
• Downloads1059
• Pages25

Abstract

Machine learning (ML) is one of the technologies that are being utilized in the context of artificial intelligence (AI), and it is becoming more significant in the healthcare industry. In its core, machine learning algorithms that detect patterns and rules with increased "feeding" of appropriate digital health data and build up an ML model following a relevant learning phase constitute the basis of this. In medicine, there are several applications of machine learning, ranging from automated diagnostic techniques to forecast of illness development to therapeutic ideas, all of which may take into consideration different features such as, for example, severity or subforum’s of diseases.

Introduction

Get free samples written by our Top-Notch subject experts for taking online Assignment Help services.

There are several potential benefits for patients and practitioners, including a reduction in waiting time and hence earlier application of "correct" treatment, which may result in the slowing down or prevention of a disease or even the possibility of recovery; similar results can also be utilized. B. to aid in the formulation of diagnostic and treatment recommendations. Doctors gain from improved assistance and safety while making treatment choices. It is also beneficial to the advancement of therapies to make predictions about which patient groups will benefit from a certain therapy or which subsequent dangers will occur with certain constellations of traits. 

In terms of the healthcare system, this leads in cheaper expenditures because of more focused therapies being administered from the beginning. Pharmaceutical companies use corresponding findings, for example, when conducting clinical trials, to select trial sites in a more targeted manner, which is particularly relevant forever smaller patient populations and individual therapies. Pharmaceutical companies also use corresponding findings when conducting clinical trials. Companies will also benefit from this since it will save them time and money. 

The use of artificial intelligence (AI) in the healthcare sector is now growing at a rapid pace, with projections indicating that revenues will more than double because of the deployment of AI-based technologies in the next years. This demonstrates that people have high expectations of the technology. Predictive analytics is a technique that is generated from the spectrum of AI-based analytics, and it is used to determine the likelihood of the occurrence of future events in advance. 

Since collective data from daily care, also known as real-world data, IQVIA has conducted this research, which includes the study of osteoarthritis, in order to glean insights about the likelihood of disease development based on particular constellations of traits. The findings may aid in the prediction of illness development and the better tailoring of medicines to the unique course of a patient's sickness, according to the researchers.

The REAL-WORLD Problem

It is a degenerative joint condition that commonly expresses itself as start-up and load-dependent discomfort, as well as stiffness and swelling. In addition to joint effusion (active arthrosis), growing distortion of the joint and joint noise because of increasing unevenness of the cartilage surface during movement are other distinctive symptoms. Osteoarthritis is the most prevalent joint illness in the world, and it is also the most frequent joint disease, affecting over five million people. Osteoarthritis is most often seen in the knee joint, but it may also affect other joints. The treatment of osteoarthritis aims to achieve two primary objectives: pain relief and endoprosthetic replacement of the joint surface. 

Researchers researched how illness progression occurs in afflicted people and whether influencing variables play a part in this process. They found that in specifically, the goal was to determine whether individuals need an endoprosthetic replacement of the joint surface and how likely it is that they would require such a procedure in the future in these patients. Also of importance was the time gap between the first diagnosis and the determination of the requirement for joint replacement to create a prognosis regarding the period leading up to the procedure. Obtaining insight into the best treatment strategy included determining which patients should receive therapy as soon as possible to slow the further course of the disease, which patients benefit most from "watchful waiting," which sufferers should be examined and treated for any additional diseases, and which sufferers are best helped with weak or b painkillers, among other things. 

Based on the retrospective IMS Disease Analyzer database, the researchers used anonymized longitudinal diagnostic and prescription information to conduct their research. This contains sickness histories, which serve as a foundation for the identification of pattern recognition. The review includes information on treatment from 2,500 general practitioners in Germany who participated in the study. The beginning point was a pool of anonymized data from more than two million osteoarthritis patients, which served as the basis for the study. After going through a multi-stage criterion selection process, about 48,000 patients were left with treatment information that could be included in the study (E, 1891). Following the creation of diverse patient profiles by cluster analysis, different disease progressions were computed using machine learning methods. 

Project Objectives

According to the findings of the AI study, the age of patients at the time of their first diagnosis, as well as their multimorbidity, which includes past infections, have a significant influence on the course of the illness. For example, those over the age of 68 who have more than six concurrent disorders may need joint replacement as soon as three years after being diagnosed with the first ailment (Schwalbe & Wahl, 2020). Joint replacement is only required after about 10 years in people who are younger and have had no or just a few infections. The breadth of treatment with general analgesics, as well as with low- and high-potency opioids, was shown to be important in addition to these key influencing variables (Gavhane et al., 2018). The findings allow for the identification of individuals with the highest medical need and the formulation of treatment recommendations that are adequately optimized. They also make it possible to optimize clinical studies by carefully deciding on the inclusion and exclusion criteria for patients before they begin (Wu et al., 2019).

Adopted Artificial Intelligence Approach

Will artificial intelligence (AI) disempower physicians in the medium to long term? Therefore, if it is feasible to make the findings of AI analyses accessible in the doctor's office software in a straightforward way, the practitioners would be assisted and freed of their burden (Liang et al., 2019). The attending physician will continue to make the ultimate decision on the course of treatment. Furthermore, the prediction of illness development represents only a tiny portion of the whole range of probable outcomes. "It is also possible to apply artificial intelligence analyses (Krittanawong et al., 2018). B, for example, may be used to create predictions regarding non-adherence, which is well-known to be a significant factor in the failure of treatments. 

It is possible to get insight into the causes for non-adherence and to take necessary efforts to improve them if you have this information. It is also feasible to determine the ideal dose, which may be important for adherence as well as compliance "Simpler expresses himself in this way: Current artificial intelligence approaches do not yet represent ordinary practices, and several legal problems, such as responsibility and data protection, still need to be defined based on the field of application in which they are used. This indicates that growth will be steady rather than unpredictable. Although the availability of suitable "big data," new technology, and analytical talent is expanding, the prospects to dramatically enhance healthcare are becoming more limited (Schmidt-Erfurth et al., 2018).

Evaluation, Results and Discussions

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

train_data = pd.read_csv("Training.csv")
test_data = pd.read_csv("Testing.csv")

train_data.head()

   itching  skin_rash  nodal_skin_eruptions  continuous_sneezing  shivering  \

4                   0        0                     0                  0  

          prognosis  Unnamed: 133 
0  Fungal infection           NaN 
1  Fungal infection           NaN 
2  Fungal infection           NaN 
3  Fungal infection           NaN 
4  Fungal infection           NaN 

[5 rows x 134 columns]

train_data.info() 

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4920 entries, 0 to 4919
Columns: 134 entries, itching to Unnamed: 133
dtypes: float64(1), int64(132), object(1)
memory usage: 5.0+ MB

train_data.isnull().any()

itching                 False
skin_rash               False
nodal_skin_eruptions    False
continuous_sneezing     False
shivering               False
                        ... 
blister                 False
red_sore_around_nose    False
yellow_crust_ooze       False
prognosis               False
Unnamed: 133             True
Length: 134, dtype: bool

train_data.drop(["Unnamed: 133"], axis = 1, inplace = True)

train_data

[4920 rows x 133 columns]

X_train = train_data.iloc[:, :-1]
X_test = test_data.iloc[:, :-1]
y_train = train_data.iloc[:, -1:]
y_test = test_data.iloc[:, -1:]

print("X train shape: ",X_train.shape)
print("y train shape: ",y_train.shape)

X train shape:  (4920, 132)
y train shape:  (4920, 1)

print("X test shape: ",X_test.shape)
print("y test shape: ",y_test.shape)

X test shape:  (42, 132)
y test shape:  (42, 1)

x = np.array([1,23,5,564,56,876,7,-123])

standardized_X = (x - np.mean(x)) / np.std(x)
normalized_X = (x-np.min(x) / np.max(x) - np.min(x))
print("Standardized array: ",standardized_X)
print("Normalized array: ",normalized_X )

Standardized array:  [-0.53531619 -0.46806733 -0.52308913  1.18564322 -0.36719405  2.13935429
-0.5169756  -0.91435521]
Normalized array:  [1.24140411e+02 1.46140411e+02 1.28140411e+02 6.87140411e+02
1.79140411e+02 9.99140411e+02 1.30140411e+02 1.40410959e-01]

from sklearn.preprocessing import StandardScaler
stds = StandardScaler()
x = x.reshape(-1,1) 
x = stds.fit_transform(x)

class NeuralNet(object):
    def __init__(self):
        np.random.seed(1)
       
        self.synaptic_weights = 2 * np.random.random((3, 1)) - 1
       
    def _sigmoid(self,x):
        return 1 / (1 + np.exp(-x))
   
    def derivative_sigmoid(self, x):
        return x * (1 - x)
   
    def train(self, inputs, outputs, iteration_number):
        for iteration in range(iteration_number):    
       
            output = self.learn(inputs)
       
            error = outputs - output
      
            factor = np.dot(inputs.T, error * self.derivative_sigmoid(output))
            self.synaptic_weights += factor
       
    def learn(self, test_inputs):
        return self._sigmoid(np.dot(test_inputs, self.synaptic_weights))

neural_net = NeuralNet()
inputs = np.array([[0, 1, 1], [1, 0, 0], [1, 0, 1]])
outputs = np.array([[1, 0, 1]]).T
neural_net.train(inputs, outputs, 50)
test_inputs = np.array([1, 0, 1])
threshold = 0.5
if neural_net.learn(test_inputs) >= threshold:
    print("Our test example output is: 1")
else:
    print("Our test example output is: 0")

Our test example output is: 1

y_train_dum = pd.get_dummies(y_train)
y_train_dum

      prognosis_(vertigo) Paroymsal  Positional Vertigo  prognosis_AIDS  \

[4920 rows x 41 columns]

import tensorflow as tf
from tensorflow.keras.models import Sequential  
from tensorflow.keras import layers 
from tensorflow.keras.layers import Dense

classifier = Sequential()

classifier.add(Dense(64, activation = "relu", input_dim = X_train.shape[1]))
classifier.add(Dense(32, activation = "relu"))
classifier.add(Dense(y_train_dum.shape[1], activation = "softmax"))

classifier.compile(optimizer = "adam", loss = "categorical_crossentropy", metrics = ["accuracy"])
classifier.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                Output Shape              Param #  
=================================================================
dense (Dense)               (None, 64)                8512     
                                                               
dense_1 (Dense)             (None, 32)                2080     
                                                               
dense_2 (Dense)             (None, 41)                1353     
                                                               
=================================================================
Total params: 11,945
Trainable params: 11,945
Non-trainable params: 0
_________________________________________________________________

history = classifier.fit(X_train, y_train_dum, epochs = 5, batch_size = 30)

Epoch 1/5
164/164 [==============================] - 1s 1ms/step - loss: 2.3234 - accuracy: 0.6187
Epoch 2/5
164/164 [==============================] - 0s 1ms/step - loss: 0.1719 - accuracy: 0.9990
Epoch 3/5
164/164 [==============================] - 0s 1ms/step - loss: 0.0279 - accuracy: 1.0000
Epoch 4/5
164/164 [==============================] - 0s 1ms/step - loss: 0.0119 - accuracy: 1.0000
Epoch 5/5
164/164 [==============================] - 0s 1ms/step - loss: 0.0067 - accuracy: 1.0000

prediction = classifier.predict(X_test)
prediction

array([[1.74469496e-05, 6.60899886e-08, 1.91234881e-06, ...,
        1.15525008e-05, 2.60058573e-06, 3.98446275e-07],
      [3.48704889e-05, 1.03426281e-08, 2.25916051e-06, ...,
        6.22341929e-07, 1.08463191e-05, 3.61527910e-08],
      [1.09912187e-04, 1.03347354e-07, 1.67949509e-06, ...,
        5.56675595e-08, 5.98589764e-08, 5.68072291e-11],
      ...,
      [1.13509273e-07, 1.61471835e-04, 5.20171525e-05, ...,
        7.80744449e-06, 1.01241812e-06, 7.34280619e-08],
      [5.95711526e-07, 1.89063241e-04, 1.12966547e-04, ...,
        3.19446553e-06, 1.94461450e-06, 7.59893695e-08],
      [6.37489793e-05, 1.32050468e-02, 2.21887007e-02, ...,
        1.09384244e-03, 2.42198061e-04, 1.08137810e-05]], dtype=float32)

history.history["accuracy"]

[0.6186991930007935, 0.9989837408065796, 1.0, 1.0, 1.0]

import matplotlib.pyplot as plt
plt.plot(history.history["accuracy"])
plt.title("Model accuracy")
plt.xlabel("Epoch")
plt.ylabel("Accuracy score")
plt.show()

plt.plot(history.history["loss"])
plt.title("Model loss")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.show()

Conclusion and Future Work

The use of artificial intelligence around cardiology has the potential to provide significant benefits. Using magnetic resonance imaging scans of the heart, for example, researchers at Johns Hopkins University demonstrated how specifically trained algorithms may be used to predict the likelihood and timing of a probable heart attack. The research also included information on the individuals who were impacted, such as their age, weight, and medications they were taking. Results showed that the predictions and computations of the algorithm were more exact than the evaluations made by doctors in this case. 

The researchers believe that by utilizing artificial intelligence to calculate risk, attending doctors will be able to make better and more effective judgments for individuals suffering from heart disease in the future. AI may potentially be able to anticipate the likelihood of developing a lethal malignancy. Researchers from Harvard University and the University of Copenhagen, for example, presented their findings at the American Society for Cancer Research's annual conference this year. Your artificial intelligence (AI) understands the individual risk of pancreatic cancer. This form of cancer is particularly hazardous since the symptoms generally manifest themselves after it is too late to have surgery. It is estimated that around 10,000 persons in Germany were diagnosed with pancreatic cancer in 2018. 

This figure comes from the Robert Koch Institute. For around 9,000 of them, assistance arrived much too late. It becomes even more critical to obtain an accurate risk estimate or a cancer diagnosis that is especially timely in this situation. As part of its efforts to advance Alzheimer's research, the Japanese corporation Fujifilm, in collaboration with Japan's National Center of Neurology and Psychiatry, is investigating the potential of artificial intelligence. It is the issue of whether Alzheimer's disease may evolve from moderate cognitive impairment during a two-year period that is the focus of the study. Researchers and others who may be impacted must first wait for the findings of more clinical studies before using artificial intelligence to routine medical practice for diagnosis and risk assessment. However, academics have already agreed that the potential of artificial intelligence is vast and that it could play a significant role in healthcare choices in the future provided it is properly implemented.