Second Pass Paper 1

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Research Methodology Assignment

Second Pass to Research paper #1

Submitted by : FA23-RCS-006
Submitted to : Sir Kashif Bilal
Date of Submission : June 12, 2024
QuCNet:- Quantum- Inspired Convolution Neural
Networks for optimized Thyroid nodule classification
Abstract:-
• Challenges addressed by this paper
Primarily attributed to the constrained datasets size for medical images and laborious
process of feature selection.
• Proposed approach
1. Quantum filter transformation for feature extraction
2. Classical Neural Network is used for classification of Thyroid nodules in
ultrasound images
• Results of Tumor Classification
→ 97.63% (Accuracy)
→ 97.72% (Precision)
and recall of 97.64% on a test dataset containing 127 images
• For Suspicious Level Classification:-
→ 93.87% (Accuracy)
→ 94.58% (Precision)
and recall of 93.88 % on a test dataset containing 98 images

Data Set Used:-


Digital database of Thyroid Ultrasound Images.

Conventional approach to detect Tyroid Nodule as benign or


mallignant:-
Tyroid Imaging Reporting and Data System(TIRADS) approach:-
• Doctors use Ultrasound to examine the lump looking at features like size, shape
and internal makeup.
• Each features gets points based on how suspicious it look.
• All the points are added up to give the lump TI-RADS Score
• Higher score means the lump looks more suspicious.

Traditional Machine Learning Models Employed:-


• Support Vector Machine (SVM)
• Naive Bayes
• Logistic Regression
• Decision Trees
Challenges: Struggle to capture intricate and Complex pattern
Deep Learning Models:-
• Pre-trained CNNs
• Transformer Models
• Attention Mechanism
Challenges :- the models have limited generalization and also time consuming feature
selection from small ultrasound dataset

Challenges to address with Traditional Models:-


Small datasets:-There often aren’t enough medical images available, especially for rare
disease. Traditional methods need a lot of data to work well, and with limited data, they
might struggle to learn the right features.
Complex images:-Medical images are complex. Traditional methods might miss these
details or get overwhelmed by the complexity.
Time and Resources:- traditional methods can be very time-consuming and require a
lot of computing power.

Quantum inspired Convolution Neural Network:-


• CNNs are powerful tools used in image recognition, where they analyze image
pixels by pixels to identify objects
• Regular CNNs used real numbers for their filters.
• QICNNs take inspiration from Quantum Mechanics and use complex number
instead

Primary Contribution of this paper:-


• A diagnostic system for thyroid cancer that utilizes Quantum inspired CNNs
• The proposed model utilizes a quantum filter transformation, extracting four
channel Quantum features from images
• This proposed model, proves to be highly effective for tumor classification in
thyroid ultrasound images

You might also like