Diagnostically Lossless Compression of Pathology Image Slides
Alexis B. Carter ; Emory University;
Content:
Image compression enables efficiencies in the storage and communication of digital pathology content. While mathematically lossless compression provides a guarantee of high fidelity, pervasive telepathology may depend on a higher level of compression, for which we propose the guiding criterion of diagnostic losslessness (DL).
Technology:
A SPOT Insight" Digital Camera was used to capture 8 uncompressed images from slides of morphologically distinct lesions. MATLAB was used to perform JPEG compression on each digital photomicrograph. Blackboard Academic Suite was used to create and administer blind evaluations of images by 11 surgical pathologists and 8 pathology residents with at least 5 months of surgical pathology training.
Design:
The subjective test was divided into 4 sessions. Each session included 2 different sets, with each set representing 5 differently compressed images. Each set was presented twice, with its individual stimuli in random order. Images were rated using an opinion scale: Acceptable, Somewhat Acceptable, or Unacceptable. The ratings were converted to a numerical scale of 3.0, 2.0, or 1.0, respectively. Subjects accessed the Blackboard via web browser and also provided free-format qualitative inputs. All of the subjects, except one, completed the sessions on a single computer of their choice.
Results:
An empirical threshold of 2.5 or higher was equated to DL. Of the 8 test images, 4 images (Anaplastic Thyroid Carcinoma, Brain Blastomycosis, Paraganglioma, Warthins Tumor) achieved DL at one or more of the milder compression levels in the range (15:1 to 43:1). The uncompressed photomicrograph represented a digital rate of 24 bits per pixel. Poor color contrast, defocusing, and pixelation were common amongst the feedback for low ratings. The remaining four images (Ovarian Yolk Sac Tumor, Pilar Cyst, Endometrial Cystic Atrophy, Uterine Leiomyosarcoma) failed the DL criterion even at the mildest compression level.
Conclusion:
Based on the observed consistency in subjective inputs, the DL criterion seems to be a viable tool in optimizing image compression for pathology slides. Television-based or teleconferencing-based compression ratios of 20:1 to 50:1 seem to be too severe as a general rule for Telepathology. A compression ratio of 10:1 seems to be a pragmatic and image-independent criterion to be validated in future testing.
