Method and analysis devices for classifying tissue samples
A method and analysis devices for classifying tissue samples are provided. In the method, annotated training data is generated from a known positive tissue sample and a known negative tissue sample from a patient, and the annotated training data is then used to train an automatic classifier patient-specifically. To delimit an affected tissue region of the patient, then, unknown tissue samples from the same patient are classified by the automatic classifier trained for the patient.
1. A method for classifying tissue samples, the method comprising:
generating annotated training data from at least one known positive tissue sample and at least one known negative tissue sample, the at least one known positive tissue sample having a specified property, the at least one known negative tissue sample not having the specified property, wherein the at least one known positive tissue sample and the at least one known negative tissue sample originating from a same patient,
providing the annotated training data to an automatic classifier;
identifying, for the same patient from whom the at least one known positive tissue sample and the at least one known negative tissue sample originate, whether a tissue sample from the same patient has the specified property, the identifying comprising training the automatic classifier patient-specifically using the annotated training data;
delimiting a tissue region affected by the specified property from a plurality of unknown tissue samples originating from the same patient, it not being known whether the plurality of unknown tissue samples have the specified property, the delimiting comprising generating input data for the automatic classifier; and
automatically classifying, by the automatic classifier, the plurality of unknown tissue samples with respect to the specified property using the input data.
2. The method of claim 1 , wherein the automatic classifier comprises adaptable parameters,
wherein the method further comprises processing the input data, the processing of the input data comprising adjusting the adaptable parameters in dependence on the annotated training data.
3. The method of claim 2 , wherein the adaptable parameters comprise a neural network, a non-linear polynomial function, a fuzzy logic system, or any combination thereof.
4. The method of claim 2 , wherein adjusting the adaptable parameters comprises automatically adjusting the adaptable parameters.
5. The method of claim 2 , wherein the adaptable parameters are pretrained using an amount of annotated training data from a number of different patients and post-trained using the at least one known positive tissue sample and the at least one known negative tissue sample patient-specifically for the respective patients from whom the plurality of unknown tissue samples for subsequent classification originate.
6. The method of claim 2 , wherein the adaptable parameters are exclusively trained using training data generated in each case from a plurality of known positive and known negative tissue samples from the same patient from whom the plurality of unknown tissue samples for subsequent classification originate.
7. The method of claim 1 , wherein to generate the annotated training data and the input data, the respective tissue samples are detected optically, and the annotated training data and the input data are analyzed by image processing for classification by the automatic classifier.
8. The method of claim 1 , wherein generating the annotated training data comprises:
processing a tissue sample from a central region of a tumor as a known positive tissue sample of the at least one known positive tissue sample;
processing a tissue sample from an environment of the tumor as a known negative tissue sample of the at least one known negative tissue sample; and
processing tissue samples for which sampling locations have a shorter spatial distance to a sampling location of the known positive tissue sample than a sampling location of the known negative tissue sample as the plurality of unknown tissue samples.
9. The method of claim 1 , further comprising:
providing, by an electronic device, an image of the affected tissue region and an environment of the affected tissue region with respective image coordinates of sampling locations of classified tissue samples to an analysis device;
automatically ascertaining, by the analysis device, an estimated extent of the affected tissue region based on the respective image coordinates; and
automatically visualizing, by the analysis device, the estimated extent in the image.
10. The method of claim 1 , further comprising:
providing, by an electronic device, an image of the affected tissue region and an environment of the affected tissue region with respective image coordinates of sampling locations of classified tissue samples to an analysis device;
automatically ascertaining, by the analysis device, a suggestion for a sampling location for a further unknown tissue sample for delimiting the affected tissue region based on the respective image coordinates; and
outputting the suggestion to a user.
11. An analysis device for classifying tissue samples, the analysis device comprising:
an input interface configured to receive annotated training data generated from tissue samples comprising at least one known positive tissue sample and at least one known negative tissue sample, the at least one known positive tissue sample having a specified property, the at least one known negative tissue sample not having the specified property, wherein the at least one known positive tissue sample and the at least one known negative tissue sample originating from a same patient;
an automatic classifier configured to identify, for the same patient, whether a tissue sample from the same patient has the specified property, therein training the automatic classifier patient-specifically using the annotated training data; and
an output apparatus configured to output corresponding respective classifications of the tissue samples,
wherein the analysis device is configured to provide the annotated training data to the automatic classifier,
wherein the analysis device is further configured to delimit a tissue region affected by the specified property from a plurality of unknown tissues samples originating from the same patient, it not being known whether the plurality of unknown tissues samples have the specified property, wherein the delimiting comprises a generation of input data for the automatic classifier, and
wherein the automatic classifier is configured to automatically classify the plurality of unknown tissue samples with respect to the specified property using the input data.