ITS70304 Principles of AI - 2022 Semester - Final Assessment: Children Autism Spectrum Disorder Screening with AI
FINAL ASSESSMENT MASTER OF APPLIED COMPUTING MARCH 2022 SEMESTER (BLOCK 2)
Case Study — Children Autism Spectrum Disorder Screening with AI
Autistic Spectrum Disorder is a major psychiatric disorder (ASD). ASD is a mental disorder that limits the use of linguistic, communicative, cognitive, and social skills. It is estimated that one out of every 100 children in the world has autism (Zeidan et. al., 2022). This estimate is an average, and reported prevalence varies greatly between studies. ASD affects a wide range of people in terms of symptoms and IQ. Traditionally, it was assumed that approximately 45 percent of autistic people are nonverbal or have intellectual disability
(ID), while the remaining 55 percent have an IQ in the average or higher range.
However, recent research suggests that the rates of co-occurring conditions are changing, with ID or language delay only being seen in 20-30% of autistic people. Autism is also more
common in men, with a male-to-female ratio of 3:1. However, there is evidence that autistic females may be underdiagnosed or receive a later diagnosis due to differences in symptom presentation and symptom camouflage (Weir et. al., 2020).
In the behavioural sciences, ASD has recently been studied using AI models to shorten screening time or improve sensitivity, specificity, or accuracy of the diagnosis process. These models are intended to be fed into a screening tool in order to achieve one or more of
the previously mentioned goals. Figure 1 depicts the English versions of the Autism Spectrum Quotient (AQ-10) for children and adolescents, which was developed to assess autistic traits. This case study is based on data from ASD screening for children. Table 1 displays a sample record from the dataset. The complete dataset (ASD_child.csv) is available on TIMeS.
1. Analyse the given full dataset and complete the following tasks: (30 marks)
a) Perform the preliminary tasks below in Python (Google Colab) (10 marks)
- Import dataset and libraries. Explain each libraries functions (6 marks)
- Find number of participants with ASD classification = NO (2 marks)
- Find number of participants with ASD classification = YES (2 marks)
b) Clean the dataset using pre-processing techniques in Python (Google Colab) (20 marks)
- Check for imbalanced in target class (5 marks)
- Data cleansing and pre-processing (10 marks)
- Split data into train and test set based on Pareto Principles (5 marks)
2. Reconstruct a piece of programming in Python (Google Colab) to develop a neural network model that predicts a child's likelihood of having autism. You are allowed to amend the data set (justify your amendments). (20 marks)
- Define and build a neural network model (10 marks)
- Train the neural network model (5 marks)
- Neural network model prediction (5 marks)
3. Based on the original dataset (or your updated dataset), reconstruct a piece of program in Python (Google Colab) to conduct a regression modelling. (20 marks)
- Define and build a regression model (10 marks)
- Train the regression model (5 marks)
- Regression model prediction (5 marks)
4. Imaging and electronic medical records (EMR), laboratory diagnosis, preventive and precision medicine, biological extensive data analysis, speeding up processes, data storage and access for health organisations are all examples of AI applications in healthcare that have literally changed the medical field. For example, in the case study presented, allowing AI models to predict the likelihood of having autism allows healthcare professionals to prioritise their resources. When it comes to predicting and/or identifying causes of illness, the accuracy of AI models is the most important factor that makes such developments successful and reliable. The more accurate the models, the more precise the results in various scenarios, and the more relevant such a model is for use in real life and as a healthcare system solution.
Based on the models developed in question 2 and 3: (30 marks)
- a) Evaluatetheperformanceofbothmodelsusingtheperformancemetric(10marks)
- b) Discuss how machine learning can be utilised by entrepreneurs to develop
technology-driven models to provide solutions for healthcare system. (10 marks)
- c) Discuss the data privacy and ethical issues associated with integrating artificial
intelligence into the healthcare system. (10 marks)
— END OF QUESTION PAPER —
