The Dawn of AI in Oncology: Transforming Cancer Prediction, Diagnosis, and Prognosis

Cancer. A two-syllable word that for most people will evoke a sharp pang of pain. Most frequently, mortuary and disturbed scenes come to mind next. Why? For ages, humanity has been beset by this pathosis, which has resulted in an immense amount of death and misery. Despite tremendous advancements in research, cancer remains the second leading cause of death worldwide, taking the lives of 10,000,000 people year. After being first used in medicine in the 1970s and enduring numerous "winters," artificial intelligence has advanced oncology to the forefront. Numerous studies have demonstrated how AI can be incorporated into the realm of oncology. Given the huge burden associated with cancer diagnosis and treatment on the healthcare system, AI has a significant chance to take the lead in risk prediction and prevention. Although there are certain risk factors that we are aware of, they are sadly only broad indicators that could be the cause (obesity, HPV). We might be able to predict the likelihood of developing cancer via AI. Target screening and the effective and efficient application of early interventions could be possible by this. The patient's medical history up until the diagnosis can be used as a data-stream to input into an algorithm. Additionally, there is a technique for identifying characteristics in provided data to generate predictions (supervised learning). An example is distinguishing between benign and malignant tumours in CT and MRI scans, which radiologists would overlook. Clinical screening is another area that is as important for risk prediction. This can be combined with AI to input and evaluate the data and decide if further diagnostic procedures or tests are needed. This is accomplished by feeding a raw image pixel into a convolutional deep learning neural network, which is trained using techniques derived from radiologist labelled outputs. This can be observed in narrow task models, which are designed to locate lesions and determine the likelihood of malignancy based on their scans, such a mammography for breast cancer or CT scan for lung cancer. It was discovered that when they were used in the medical field, their screening analysis was equivalent or superior to that of skilled diagnosticians. Radiologists could then concentrate on other areas that call for their expert judgement because AI can analyse images quickly and effectively. The prostate specific antigen (PSA) machine can be trained algorithms that model past various tie points in conjunction with other serum indicators, which is another method artificial intelligence is applied in serum markers. Because of this, the machine can identify prostate cancer more accurately than a PSA by itself. It's crucial to remember that while these algorithms show excellent results in terms of sensitivity and specificity, they are unable to assess direct clinical outcomes like quality of life or cancer mortality. This intelligibly implies a field of application for AI. Risk assessment and prognosis represent yet another promising application area for AI. Risk categorisation was formerly based on TNM staging. This concept attempts to make a prognosis by first analysing the cancer physically. This means that it is unable to take into account the various patient, tumour, and environmental spectra that may have an impact on the prognosis. Adding data streams (genomics, serum markers, and sophisticated imaging) can enable more accurate risk categorisation through AI. One instance is the development of machine learning-based genomic classifiers for malignant growths. This is demonstrated by the logistic regression-based classifier OncotypeDx for breast cancer, which has the ability to guide treatment decisions and enhance prognosis. Furthermore, employing combinations of genomic data, EHR data, and diagnostic imaging, deep learning algorithms have been researched to integrate multi-omic data sources into risk classification. Despite recent years of great research in clinical oncology, there is a significant discrepancy between the evidence supporting therapeutic impact and performance. Only two published randomised clinical trials of deep learning in medical imaging were found in a recent review, despite thousands of published research on deep learning logarithm performance. The main obstacle to AI development appears to be the lack of sufficient data, both in terms of number and quality. The successful development of AI and its integration into the medical profession, making it accessible to all, requires the resolution of this and other challenges.