Role Of AI/ML In Improving Treatment Outcomes By Optimizing Healthcare Workflows
Artificial Intelligence can optimise diagnostic workflows by augmenting non-invasive investigative modalities. Histopathological examination of biopsy tissue with ancillary procedures, is considered the gold standard for cancer diagnosis.
In the last decade, significant advances in computational pathology, have led to increased applications of Artificial Intelligence/Machine Learning (AI-ML) in diagnostic Oncopathology. The current crop of AI-ML pathology solutions help augment cancer diagnostics with applications through the entire care pathway of screening, diagnosis, prognostication & treatment. They are making a meaningful impact on different aspects of the cancer care pathway in a couple of ways:
Improving efficiency of diagnostic workflows
AI applications can impact the cost, resource utilisation and turnaround time of diagnostic workflows. One example is 'triaging' or 'sorting' applications, to optimise laboratory workflows.
Diagnostic firms are training networks for triaging tissue images from prostate cancer patients. These networks can separate images positive for cancer from the negative ones and improve the speed of screening tissue images. They also reduce potential errors of omission and help prevent missed diagnosis of cancer.
Another example is solutions for the basic but crucial task of assessing adequacy of bronchoscopic lymph node aspirates from lung cancer patients.
In a conventional workflow, a pathologist needs to be available during the aspiration procedure for carrying out a real-time examination of the aspirates in order to confirm their adequacy for diagnosis. As one can imagine, this places a large coordination burden on the simultaneous availability of pathologist and pulmonologist. Not having both can create a scenario where the patient has the unnecessary burden of needing to undergo a repeat procedure. This leads to delay in reporting, not to mention the added expense and morbidity due to the repeated invasive procedure. The new algorithm for on-site assessment provides immediate adequacy results ensuring that the patient is not subjected to a second hospital visit for another bronchoscopy.
Another important way in which Artificial Intelligence can optimise diagnostic workflows is by augmenting non-invasive investigative modalities. Histopathological examination of biopsy tissue with ancillary procedures, is considered the gold standard for cancer diagnosis. However this modality is invasive and time consuming. In prostate cancer, radiological imaging tests are often used as first line of investigation to detect a malignancy which is then biopsied for confirmation by histopathological examination. But imaging techniques like MRI substantially underestimate the size, extent and tumor volumes in comparison with histopathology.
To surmount both problems, diagnostic companies are training networks on histopathology images from prostate surgeries and translating this learning to improve accuracy of tumour identification and tumour volume estimation in mpMRI images.
Prognostication, prediction and treatment selection for Cancer
AI-ML algorithms are effective at accurately quantifying different structures and cells in cancer tissue to aid cancer grading and prognostication and decide on treatment regimens for patients. These quantitative image analysis applications improve objectivity of results and promote standardised applicability of pathology findings in cancer care.
For example, Mitotic figures seen frequently in cancer cells are used to assess the aggressiveness of cancer.
The larger goal in the application of AI-ML in cancer care is to predict the course of the cancer in individual patients. This effectively paves the way for personalised healthcare - with treatment regimens tailored to individual patients compared to the current 'one size fits all' approach.