Diagnosing cancers such as pancreatic, liver, lung, kidney, ovarian, and brain cancer is one of the most prominent challenges in medicine. These cancers, often having minimal early detection signs and overlapping symptoms with other medical conditions, are often only detected when they have spread and caused considerable damage. Artificial Intelligence (AI) medical imaging technologies offer an accurate and efficient tool to detect these cancers. Through new AI technologies such as deep learning, misdiagnoses of cancer within the U.S patient population can be minimized. 

It is imperative to diagnose cancer at early stages to improve the outcome of the disease and the effectiveness of treatment. However, an inaccurate cancer diagnosis — misdiagnosis or inaccurate diagnosis — can pose a significant dilemma for the health of the patient. Misdiagnosis of cancer can be caused by human factors such as a clinician’s misinterpretation of a tissue scan and overbooked schedules. Most of these causes are completely preventable. 

A false positive diagnosis of cancer is defined as inaccurately identifying something as cancer. This can lead to the patient undergoing unnecessary procedures and biological, financial, and psychological damagedamages. In one notable case, Frank Berrera was minutes away from prostate surgery before the pathology department notified the surgical team that he did not have prostate cancer. Frank is not the only one who has experienced a false positive diagnosis. A PLoS One study estimated that the probability of receiving at least one false positive for a disease such as cancer in a lifetime is about 85.5% for men and women. 

Conversely, a false negative diagnosis of cancer (or interpreting cancer as a different medical condition) often delays treatment or gives rise to unnecessary treatment, leading to a worse prognosis. In total, about 795,000 Americans are injured or killed due to the misdiagnosis of a dangerous disease annually, including cancer. 

Human error is a major cause of cancer misdiagnosis, and AI has the potential to minimize these discrepancies in cancer detection. Deep learning is a form of AI that is able to be “trained” for pattern recognition by providing it with numerous sample photos, which it can use to identify similar features in different images. Applying deep learning to cancer detection can assist pathologists in supporting their qualitative(relating to viewing and judgment over numbers) diagnoses. AI’s ability to analyze large, complex data sets and classify cancer in a limited time can allow faster diagnosis of cancer with more accuracy and effective treatment administration. 

Is there a way to detect cancer before it occurs? With AI technology, predicting the likelihood of developing certain cancer types is possible. AI population-based models can be utilized to identify those most at risk of cancer within a population by analyzing patient data records. These early predictions of cancer can be used to allow physicians to better monitor high-risk patients and spot hard-to-detect cancers within a broad range of individuals. This is especially beneficial for cancers that are commonly detected in later terminal stages, such as pancreatic cancer. 

AI, however, is not the “perfect” program that it may seem to be;, it often makes mistakes, as with medical professionals, due to its limited generalizability. An instance of limited generalizability is a phenomenon called bias. Bias can occur in diagnosis due to discrepancies in the supplemented training population data and the different population data. Each population can have individuals with different physical, genetic, cultural, and financial aspects, which can affect how accurate an AI’s diagnosis is. If an AI tool has been trained on data from one specific population, it may provide an accurate diagnosis for an individual within that population but an inaccurate diagnosis for an individual in another. 

Errors in AI cancer diagnostic models will arise, but with consistent maintenance, correcting biases, and ensuring proper functioning, these errors can be reduced. AI, coupled with contemporary oncology, could significantly improve the accuracy of cancer diagnoses and minimize the amount of time elapsed before treatment is received. This can ultimately improve the survival of individuals with hard-to-detect cancers by allowing for accurate and immediate treatment.