AI can now detect Cancer-edited by Aishee Biswas

Dr. Éva Ambrózay, a highly experienced radiologist, scrutinized a patient's mammogram on a computer monitor at a Hospital near Budapest. Despite two previous radiologists deeming the X-ray as clear of breast cancer, Dr. Ambrózay carefully examined several red-circled areas that an artificial intelligence (A.I.) software had flagged as potentially cancerous.

 

Recent advancements in A.I. are making significant strides in breast cancer screening  by identifying signs that doctors may miss. Early results and radiologists suggest that the technology can detect cancer at least as accurately as human radiologists. This development is a concrete demonstration of how A.I. can enhance public health.

 

As the use of artificial intelligence (AI) technology expands, it is now being utilized in the detection of breast cancer. Many breast imaging centers across the country are incorporating computer-assisted detection (CAD), a type of AI, to reduce the number of missed breast cancer cases detected by traditional mammograms. A major study suggests that mammogram screenings miss about one in eight cases of breast cancer.

 

With CAD, a patient undergoes a standard mammogram, which is an X-ray of the breast, and the results are further analyzed by a computer to identify potential cancerous areas. Studies indicate that CAD helps to review images, assess breast density, and identify high-risk mammograms that may have been missed by radiologists. It can also signal that a mammogram needs to be retaken.

 

The CDC considers mammograms the most effective method for early detection of breast cancer, leading to better treatment outcomes and management for most women.

Although the application of AI in breast cancer screening is promising, it remains a developing technology. Research indicates that the use of AI in mammography may result in higher false-positive rates. To date, the most effective approach involves a combination of AI and human expertise. A recent study demonstrated that the joint effort of humans and AI can identify 2.6% more breast cancer cases with fewer incorrect diagnoses.

For brain cancer, a newly developed system showed promise in identifying mutations used by the World Health Organization to define molecular subgroups of diffuse glioma, the most common and deadly primary brain tumor. The study involved more than 150 patients and the system demonstrated an average accuracy of over 90%.

A collaboration between neurosurgeons and engineers , as well as researchers from New York University and the University of California, San Francisco, resulted in the creation of an AI-based diagnostic screening system named DeepGlioma. The system quickly analyzes tumor specimens taken during surgery and utilizes rapid imaging to detect genetic mutations.

The diagnosis and treatment of gliomas are increasingly reliant on molecular classification, as the genetic makeup of brain tumor patients affects the benefits and risks of surgery. Complete tumor removal can result in an average of five years longer survival for patients with astrocytomas, a specific subtype of diffuse glioma, compared to other subtypes.

 

Before the development of DeepGlioma, surgeons lacked a method to distinguish between various types of diffuse gliomas during surgery. Conceived in 2019, the system integrates deep neural networks with stimulated Raman histology, an optical imaging technique also created at U-M, to capture real-time images of brain tumor tissue.