Purpose:　Collagen　features　in　breast　tumor　microenvironment　is　closely　associated　with　the　prognosis　of　patients.　We　aim　to　explore　the　prognostic　significance　of　collagen　features　at　breast　tumor　border　by　combining　multiphoton　imaging　and　imaging　analysis.　Methods:　We　used　multiphoton　microscopy　(MPM)　to　label-freely　image　human　breast　tumor　samples　and　then　constructed　an　automatic　classification　model　based　on　deep　learning　to　identify　collagen　signatures　from　multiphoton　images.　We　recognized　three　kinds　of　collagen　signatures　at　tumor　boundary　(CSTB　I-III)　in　a　small-scale,　and　furthermore　obtained　a　CSTB　score　for　each　patient　based　on　the　combined　CSTB　I-III　by　using　the　ridge　regression　analysis.　The　prognostic　performance　of　CSTB　score　is　assessed　by　the　area　under　the　receiver　operating　characteristic　curve　(AUC),　Cox　proportional　hazard　regression　analysis,　as　well　as　Kaplan-Meier　survival　analysis.　Results:　As　an　independent　prognostic　factor,　statistical　results　reveal　that　the　prognostic　performance　of　CSTB　score　is　better　than　that　of　the　clinical　model　combining　three　independent　prognostic　indicators,　molecular　subtype,　tumor　size,　and　lymph　nodal　metastasis　(AUC,　Training　dataset:　0.773　vs.　0.749;　External　validation:　0.753　vs.　0.724;　HR,　Training　dataset:　4.18　vs.　3.92;　External　validation:　4.98　vs.　4.16),　and　as　an　auxiliary　indicator,　it　can　greatly　improve　the　accuracy　of　prognostic　prediction.　And　furthermore,　a　nomogram　combining　the　CSTB　score　with　the　clinical　model　is　established　for　prognosis　prediction　and　clinical　decision　making.　Conclusion:　This　standardized　and　automated　imaging　prognosticator　may　convince　pathologists　to　adopt　it　as　a　prognostic　factor,　thereby　customizing　more　effective　treatment　plans　for　patients.　©　2023,　Springer　Nature　Switzerland　AG.