Artificial Intelligence (AI) in Liver Cancer Risk Prediction: A Promising Path Towards Early Detection

Liver cancer is emerging as a major global health concern and ranks as the sixth most common type of cancer worldwide. Its insidious nature often means it remains undetected until the later stages, by which point treatment options and chances of survival can be dramatically reduced. To combat this, it is crucial to understand the risk factors associated with liver cancer and to advocate for the importance of early detection. By doing so, we can reduce morbidity rates and improve patient outcomes. Hepatitis B and C infections, a history of diabetes mellitus, exposure to aflatoxin B1, and lifestyle factors including alcohol consumption and smoking have been identified as key risk factors for liver cancer.

Artificial Neural Networks (ANNs) have experienced a surge in their application within the field of medicine over the past few years. ANNs are computational models inspired by the way biological neural networks in the human brain function. They consist of interconnected nodes or “neurons” that process information in layers. When it comes to medicine and specifically cancer risk prediction, ANNs and other machine learning techniques have shown promising results. For instance, ANNs are extensively used for image recognition tasks, and this utility extends to medical imaging as well. They can be trained to identify tumors in X-rays, MRIs, CT scans, and mammograms. Moreover, digital pathology slides can be analyzed using ANNs to detect and classify cancerous cells, thus assisting pathologists in diagnosis. Furthermore, ANNs can predict the progression of diseases and provide valuable insights into expected patient outcomes based on historical data. They can also be used to personalize treatment plans by analyzing a patient’s medical records, genetics, and other factors to determine the most effective treatment strategy. With the decreasing cost of sequencing technologies, genomic data is becoming increasingly available. ANNs can analyze these large datasets to identify genetic mutations and patterns associated with increased cancer risk. By understanding an individual’s genetic makeup, healthcare providers can potentially anticipate the types of cancers to which the individual might be susceptible, thus allowing for earlier and more targeted screenings or interventions.

A new study published in the peer-reviewed Acta Oncologica Journal led by Assistant Professor Wazir Muhammad and Dr. Afrouz Ataei, from Florida Atlantic University in collaboration with Professor Jun Deng from Yale University, successfully developed an ANN model for accurately predicting an individual’s risk of developing liver cancer. This model utilized a wide range of personal health data, including factors such as age, smoking habits, exercise habits, diabetes, BMI, and more, to make quantitative risk assessments.

In their study, the research team utilized two key datasets: the National Health Interview Survey (NHIS) data from 1997 to 2019 and the Prostate, Lung, Colorectal, and Ovarian cancer screening trial (PLCO) dataset. These datasets provided a wealth of health-related information, including age, smoking habits, exercise habits, diabetes, race, body mass index (BMI), heart disease, stroke, and more, all of which are potential risk predictors for liver cancer. To address missing data, a one-hot encoding method was employed, allowing for the imputation of missing values. The study then applied ANN models, which have demonstrated superior performance in various cancer risk prediction studies. The architecture consisted of three hidden layers with varying numbers of neurons to optimize accuracy.

The authors assessed the new ANN model’s performance using the area under the receiver operating characteristic curve (AUC) for both training and testing datasets. Notably, the AUC values for the testing dataset from the PLCO dataset exceeded 0.80, demonstrating the model’s impressive predictive accuracy. Sensitivity and specificity remained consistently high, with sensitivity hovering around 60% on the training dataset and 40% on the testing dataset, while specificity remained above 99.5%. This high AUC value indicates the model’s ability to distinguish between individuals at different levels of risk effectively.

The presented ANN model has the potential to revolutionize liver cancer risk prediction. It offers an invaluable tool for early detection by leveraging readily available electronic health records and personal health data. Traditional screening modalities, such as blood tests and imaging, are often employed when symptoms have already manifested, making early diagnosis challenging. The ANN model bridges this gap by providing a means of predicting liver cancer risk before overt symptoms appear. Previous risk prediction models for HCC have primarily focused on high-risk populations, such as those with chronic hepatitis B infection. This limited scope fails to account for lifestyle-related factors, such as physical activity, diabetes, smoking, BMI, and alcohol consumption, which can also influence an individual’s risk of developing liver cancer. The ANN model developed in the new study, however, takes a broader approach, considering data from the general population. This inclusivity makes the model applicable to a more diverse range of individuals, increasing its clinical relevance.