The ultimate aim of AI, like every advancement in medicine and pathology, is to improve both the diagnostic accuracy and lead to better outcomes for the patients. Dr Travis Brown is a general pathologist with a special interest in Information Technology and Pathology Informatics. He explains that although improvements are yet to be demonstrated in a clinical setting, the studies which prove the potential benefits are hopefully not too far away.

“At present, AI is still in the research stage. There are numerous organisations/companies around the world who are developing AI technology, however, whilst there are lots of studies which show potential benefits, this is yet to translate to routine clinical practice. Some of the many areas being researched for AI include, prostate cancer, breast cancer, gastrointestinal pathology, cytology (i.e., Cervical/PAP smears, molecular pathology (i.e., detection of BRAF V600E on H&E slides) and parasite detection (i.e., trichomonas on cervical smears).

“There are also some companies who are working on non-imaging-based AI software such as the prediction of Human Papillomavirus (HPV) infection or sepsis in patients from their medical records data,” said Dr Brown.

Machine learning (ML) is a branch of AI and computer science which focuses on the use of data and algorithms to become more accurate at predicting outcomes, enabling systems to learn and improve from experience without being explicitly programmed. It is thought that ML will provide a useful aide to pathologists in their routine work.

“A number of organisations appear to be focusing on the use of ML to diagnose certain cancers (i.e., prostate and breast), however, I believe organisations who are applying ML to areas that pathologists need assistance will be the most successful. This includes parasite detection, fungal identification, or Ki-67/mitoses quantification. If you think about it, every day pathologists are determining if a biopsy shows cancer or not so it would be rare for AI to have a significant input against this question. It is the tasks that pathologists seldom or rarely do that I believe AI will provide the most benefit,” said Dr Brown.

Whilst AI is expanding very rapidly, not every AI application is the same. A large proportion of AI systems are also user dependent meaning that features highlighted in a demonstration are because a user has shown an interest in that area.

“The challenge for laboratories and pathologists is that they will need to be aware of what it is they are purchasing when selecting AI software. In addition, refinements and learning will certainly need to be incorporated into AI systems by pathologists that will themselves become an important resource for future practice. Who owns those modifications may be an interesting discussion with the vendor.

“It will also be important for laboratories to discuss what the legal responsibilities of the AI vendors are, prior to the investment in their products, in the case of incorrect image interpretation or even misdiagnosis. Ultimately, it will be the responsibility of the reporting pathologist to sign out the case and accept responsibility for the diagnosis, taking into consideration any AI information available.

“It is important to note that the introduction of AI has an additional regulatory hurdle in its path. In Australia, AI will fall under the ‘software based medical devices’ which requires approval by the Therapeutic Goods Administration (TGA) via the Therapeutic Goods (Medical Devices) Regulations 2002. Vendors will need to prove that it is fit for ‘medical purpose’ to aide pathologists in the diagnostic process. Therefore, this is an important question for anyone interested in purchasing the latest AI for their laboratory,” said Dr Brown.