AI now classifies lung cancer as well as laboratory diagnostics

Towards healthcare with AI support

Deep learning defines cancer as good as laboratory diagnostics

In the West, lung cancer is the deadliest type of cancer. Laboratory diagnostics specialists examine tissue samples under a microscope and classify them to determine the stage of tumor development and prescribe treatment. In each case, cancer is unique in its own way, so interpreting the drug can be a difficult task. Extremely difficult. Can artificial intelligence come to the rescue?
Yes, yes, and ...

My bro John Batson Sherlock from the BBC

Deep learning

Recently, an image analysis technique has emerged among deep learning technologies, bringing major changes to the field of computer vision. It automatically identifies unique image features and is called a convolutional neural network (SNA). For automatic recognition of unique images in network images, they use an approach using data processing and perform this work better than a person if we take the ImageNet and CIFAR-10 databases marked manually as the reference. If you use a large number of virtual images of drugs with the comments of specialists, the SNA can be taught to classify various types of lung cancer according to images, and thus facilitate the process of detection and classification of lung adenocarcinoma.

Model sliding window for classifying virtual images of lung preparations

Creating AI

Scientists from the Hassanpour Lab at the Geisel School of Medicine in Dartmouth published a scientific article in Nature Scientific Reports , which talked about a neural network that can classify the histological subtypes of lung cancer: creeping, acinar, papillary, micropapillary and solid. The model was trained on more than 4,000 commented virtual slides and fine-tuned using a set of classic samples for each characteristic type. The trained model performed well on these classic samples: with an area under the curve of the operating characteristic greater than or equal to 0.97 for all categories.

AI performance indicator for classic lung cancer samples

AI vs specialists?

To compare the work of this AI and laboratory diagnostics specialists, scientists measured their performance in independent testing. The deep learning model and three medical practitioners classified 143 complete virtual images of drugs with real cases. According to the Kappa coefficient and two indicators of agreement, the model they trained bypassed the diagnostic doctors in all respects, as shown in this table from the report:

Table 2: Comparison of specialists and our model in the classification of prevailing subtypes in 143 complete virtual images of drugs. Good agreement (R. Agreement) means the consent of the commentator with at least two of the other three. 95% of confidence intervals are shown in parentheses.

To make a comparison, the characteristic types detected by the model were graphically presented slide by slide, along with experts commented on for a number of selected images. The matches are very accurate:

Visualization of the histological picture, commented by specialists ((Ai-iv) in comparison with those that determined the model of deep learning (Bi-iv).

What does it mean?

Deep learning has become an extremely powerful method that can work on a par with a person even when solving complex problems such as the analysis of medical images. A lung cancer classifier based on deep learning algorithms could divide patients into groups and prioritize cases for medical analysis. It could also serve as a second opinion in cases with obscure images. Although these methods in the future can automate the time-consuming part of the specialist’s work, much remains to be done before they can be used in practice. This model needs to be tested on many databases from different organizations. Its suitability must be confirmed by clinical trials. Is it likely that an automated system will replace laboratory diagnostics specialists? Maybe one day but not soon. All AI systems must be thoroughly tested in a clinical setting before doctors, patients, and the medical community can trust them.

Creating AI algorithms for healthcare is like climbing a high mountain. We can be halfway, but there is still a long way ahead, and it does not become easier.

The code for classifying a scan with lung histopathology is publicly available on Github .