Methods and arrangements concerning dermatology
Reference imagery of dermatological conditions is compiled in a crowd-sourced database (contributed by clinicians and/or the lay public), together with associated diagnosis information. A user later submits a query image to the system (e.g., captured with a smartphone). Image-based derivatives for the query image are determined (e.g., color histograms, FFT-based metrics, etc.), and are compared against similar derivatives computed from the reference imagery. This comparison identifies diseases that are not consistent with the query image, and such information is reported to the user. Depending on the size of the database, and the specificity of the data, 90% or more of candidate conditions may be effectively ruled-out, possibly sparing the user from expensive and painful biopsy procedures, and granting some peace of mind (e.g., knowledge that an emerging pattern of small lesions on a forearm is probably not caused by shingles, bedbugs, malaria or AIDS). A great number of other features and arrangements are also detailed.
1. A method employing a camera-equipped smartphone, comprising:
optically-capturing first data representing a part of a patient's body that evidences a symptom of a possible pathological condition, using said smartphone;
processing said first data, said processing including deriving features therefrom, said processing including applying data to an input layer of a neural network classifier, the neural network including layers of neurons having weighted connections therebetween, the weights having been determined in a supervised learning process, the neural network including an output layer of neurons that provide output data in accordance with weighted combinations of inputs thereto, based at least in part on said derived features, said features including one or more features drawn from a first list consisting of: (a) 3D skin microtopology data, (b) histogram data, (c) image data gathered under plural different spectrally tuned illumination conditions, (d) data that decomposes input imagery into plural components of different frequencies, angular orientations, phases and/or magnitudes, (e) features characterized in each of between 4 and 20 different color channels, and (f) wavelet transform data;
from the output data, determining result information, said determining including identifying plural particular pathological conditions that the neural network classifier concludes is inconsistent with the pathological condition evidenced by said depicted part of the body; and
reporting, to the patient, a name of plural of said identified pathological conditions that the neural network classifier concludes is inconsistent with the pathological condition evidenced by said depicted part of the body;
wherein the method serves to reduce patient worry that the patient may be suffering from any of said plural reported conditions.
2. The method of claim 1 in which the first imagery depicts said part of the body at a first time, and the method further includes:
receiving second imagery depicting said part of the body at a second time, later than the first time;
determining data about a change in said symptom between the first and second times, based on said first and second imagery; and
using said determined data in said identifying the plural particular pathological conditions that is not the pathological condition evidenced by said depicted part of the body.
3. The method of claim 1 that further includes determining, from said output data, plural candidate pathological conditions that are consistent with said received first imagery, and presenting a listing of said candidate conditions to the patient, ranked by probability.
4. The method of claim 1 in which the first imagery includes a known object distinct from the body, and the method includes determining a camera pose relative to the body based on apparent geometrical distortion of the object in said imagery, and applying a corrective counter-distortion to the first imagery to account for said apparent geometrical distortion.
5. A non-transitory computer readable medium containing software instructions operable to configure a system to perform acts including:
receiving first optically-captured data representing a part of a mammalian body that evidences a symptom of a possible pathological condition;
processing said first data, said processing including deriving features therefrom, said processing including applying data to an input layer of a neural network, the network including layers of neurons having weighted connections therebetween, the weights having been determined in a supervised learning process, the neural network including an output layer of neurons that provide output data in accordance with weighted combinations of inputs thereto, based at least in part on said derived features, said features including one or more features drawn from a first list consisting of: (a) 3D skin microtopology data, (b) histogram data, (c) image data gathered under plural different spectrally tuned illumination conditions, (d) data that decomposes input imagery into plural components of different frequencies, angular orientations, phases and/or magnitudes, (e) features characterized in each of between 4 and 20 different color channels, and (f) wavelet transform data;
from the output data, determining result information, said determining including identifying plural particular pathological conditions that the neural network concludes is inconsistent with the pathological condition evidenced by said depicted part of the body; and
reporting, to a user, a name of plural of said identified pathological conditions that the neural network concludes is inconsistent with the pathological condition evidenced by said depicted part of the body;
wherein the system configured by said instructions serves to reduce patient worry that the patient may be suffering from any of said plural reported conditions.
6. A mobile device including a camera, a screen, a processor, and a memory, the memory containing software instructions operable to configure the device to perform acts including:
applying imagery captured by the camera to an input layer of a neural network, the network including layers of neurons having weighted connections therebetween, the weights having been determined in a supervised learning process, the neural network including an output layer of neurons that provide output data in accordance with weighted combinations of inputs thereto;
from the output data, determining result information, said determining including identifying plural particular pathological conditions that the neural network concludes is inconsistent with a pathological condition of a mammalian body depicted in the captured imagery; and
reporting, to a user, a name of plural of said identified pathological conditions that the neural network concludes is inconsistent with the pathological condition depicted in the captured imagery;
wherein the mobile device serves to reduce patient worry that the patient may be suffering from any of said plural reported conditions.
7. The method of claim 1 wherein said features include 3D skin microtopology data.
8. The method of claim 1 wherein said features include histogram data.
9. The method of claim 8 wherein said features include a 3D histogram.
10. The method of claim 8 wherein said features include a histogram identifying frequency of occurrence of different shapes.
11. The method of claim 1 wherein said features include data that decomposes input imagery into plural components of different frequencies, angular orientations, phases and/or magnitudes.
12. The method of claim 1 wherein said features include features characterized in each of between 4 and 20 different color channels.
13. The method of claim 1 wherein said features include wavelet transform data.
14. The method of claim 1 wherein said features include one or more features computed on a first scale corresponding to a first portion of the received first data, and also includes one or more features computed on a second scale, smaller than the first scale, for each of plural second portions of the received first data that are each smaller than said first portion.
15. The method of claim 1 :
wherein said features include one or more features drawn from a smaller list consisting of: histogram data, data that decomposes input imagery into plural components of different frequencies, angular orientations, phases and/or magnitudes, and wavelet transform data; and
wherein said one or more features drawn from the smaller list are determined in each of at least five different spectral bands.
16. The method of claim 1 that further includes employing computational photography techniques to achieve an extended depth of field for data processed by the neural network.
17. The computer readable medium of claim 5 wherein said determining includes determining plural particular pathological conditions that the neural network classifier indicates are inconsistent with the pathological condition evidenced by the depicted part of the body; and the instructions are further operable to configure the system to report, to the user, the names of said plural pathological conditions.
18. The device of claim 6 wherein said determining includes determining plural particular pathological conditions that the neural network classifier indicates are inconsistent with the pathological condition evidenced by the depicted part of the body; and the instructions are further operable to configure the device to report, to the user, the names of said plural pathological conditions.
19. The method of claim 1 which includes not reporting, to the user, a name of a pathological condition that the neural network classifier concludes is consistent with the pathological condition evidenced by said depicted part of the body.
20. The method of claim 4 in which the known object is a USB plug.