Page 103 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

tools, and specifically machine learning algorithms, in the analysis of histopathology images
along with E-HRC lab at the International Institute of Information Technology – Bangalore (IIITB)

2�2 Benefits of the use case 4.1-Healthcare

Our use case pertains to this pertinent goal of good health and well-being.

2�3 Future Work

This work was carried out under the constraints of limited data sets and limited class labels.
These constraints were addressed by formulating necrosis detection as an ML problem. We
set up a complete pipeline that is seamlessly extensible. The results are encouraging, and the
work demonstrates the potential of AI/ML in Computational Neuropathology for screening
and detection of malignancies of tumours of the CNS.

Annotated data in digital histopathology is hard to acquire, unlike problems such as face
detection, which is available commonly, since analysing and labelling histopathology data
requires the involvement of expert pathologists. The availability of larger datasets will enable
further validation of the approach and enable incorporation into a tool that can be used by
pathologists. These will need to ensure adequate performance in the analysis of WSI and ensure
high rates of specificity and sensitivity across demographics and scanners, with repeatability of
results. A methodology must be defined for validation of the algorithm on a much larger set
of WSI samples, covering multiple scanners and patient demographics.

3 Use Case Requirements

• REQ-1: Need to have annotated histopathological images verified by a committee of
pathologists before proceeding to build the use case pipeline.
• REQ-2: All the artifacts, whitespace, noise etc must be removed before further processing.
• REQ-3: Augmentation needs to be performed depending on the dataset size.
• REQ-4: Handcrafting of features must be done before the Machine Learning and
Ensemble Algorithms are applied.

67